dune far detector task force preliminary report
TRANSCRIPT
DUNE Far Detector Task Force Preliminary Report
September 30, 2016
Contents
1 Introduction 2
2 Simulation and Reconstruction Chain 32.1 Monte Carlo Challenge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Simulation Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.1 Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2.2 geant4 Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.3 Digitization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Reconstruction Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.3.1 Signal Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.3.2 Gaussian Hit Finder . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.3.3 Disambiguation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.3.4 Line Cluster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.3.5 Blurred Cluster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.3.6 Pandora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3.7 Projection Matching Algorithm . . . . . . . . . . . . . . . . . . . . . 102.3.8 EMShower . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.3.9 Calorimetric Energy Reconstruction and Particle Identification . . . . 122.3.10 WireCell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.3.11 Optical Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Dual Phase Simulation and Reconstruction Chain 17
4 Non-beam physics 214.1 Supernova and Low-Energy Neutrinos . . . . . . . . . . . . . . . . . . . . . . 21
4.1.1 Nature of the Signal and Reconstruction Challenge . . . . . . . . . . 214.1.2 SNOwGLoBES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234.1.3 Event Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244.1.4 Report from Hack Days . . . . . . . . . . . . . . . . . . . . . . . . . 254.1.5 Reconstruction Progress . . . . . . . . . . . . . . . . . . . . . . . . . 254.1.6 DAQ and Triggering Progress . . . . . . . . . . . . . . . . . . . . . . 274.1.7 Photon Studies for Low-Energy Events . . . . . . . . . . . . . . . . . 28
1
4.2 Nucleon Decay Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284.3 Atmospheric Neutrino Physics . . . . . . . . . . . . . . . . . . . . . . . . . . 364.4 Cosmic Rays and Cosmogenics . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.4.1 Muon generator: MUSUN . . . . . . . . . . . . . . . . . . . . . . . . 394.4.2 Muon background for proton decay . . . . . . . . . . . . . . . . . . . 434.4.3 Cosmic-ray event reconstruction . . . . . . . . . . . . . . . . . . . . . 464.4.4 Cosmogenics and Requirements for the Far Detector . . . . . . . . . . 48
5 Long-baseline Physics Sensitivity Studies 495.1 Event Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495.2 Oscillation Fits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2.1 VALOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515.2.2 LOAF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6 Detector Optimization Studies 556.1 Wire Spacing and Wire Angle Studies . . . . . . . . . . . . . . . . . . . . . . 55
6.1.1 Tracking Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 556.1.2 e/γ Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.2 Perpendicular vs. Parallel APA Comparison . . . . . . . . . . . . . . . . . . 686.3 Considerations of implementing an additional readout wire plane in the single-
phase LArTPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 716.4 Photon Detector System Task Force . . . . . . . . . . . . . . . . . . . . . . . 74
7 Summary and Next Steps 75
1 Introduction
The DUNE Far Detector Task Force was formed in September 2015 and charged to:
• develop a full far detector (FD) simulation and reconstruction chain (Section 2);
• produce detector optimization studies, for example, 3 mm vs 5 mm wire pitch, wireangle and the efficiency of the light readout system for different configurations (Sec-tion 6);
• produce a first update the DUNE long-baseline physics sensitivity studies using fullsimulation and event reconstruction (Section 5);
• develop the simulation and reconstruction for SNB and nucleon decay physics (Sec-tion 4);
• take into account both reference (single-phase TPC) and alternate (dual-phase TPC)FD designs in the studies (Section 3).
These goals were motivated by the status of DUNE’s simulation and reconstruction toolsand physics sensitivity studies at the time the task force was formed. The long-baseline oscil-lation sensitivity studies up until that time had been performed with GLoBES [1, 2]. GLoBES
2
is not a full Monte Carlo (MC) simulation, but takes neutrino beam fluxes, cross sections,and detector-response parameterization as inputs to produce simulated energy spectra thatare used in a fit for oscillation parameters. Simulations of neutrino beam events in DUNEhad been performed, but reconstruction tools were still in the very early stage of devel-opment. Tools to simulate nucleon decay, atmospheric neutrinos, and supernova neutrinoshad not yet been incorporated into the software. A full simulation and reconstruction chainis needed to study how changes in detector design affect the final sensitivities. The goalof this task force is to develop simulation and reconstruction tools necessary to study theeffect of detector design on physics performance. We have selected a few particular detectordesign questions to address, but we expect detector optimization studies to continue afterthe conclusion of the task force.
This document represents the preliminary report on the findings of the Far Detector TaskForce. A final report will be presented in March 2017.
2 Simulation and Reconstruction Chain
In this section, we summarize the latest development of the simulation and reconstructiontools that are being used by the DUNE 35t prototype, far detector, and protoDUNE. Theadoption of the common framework LArSoft [3] by several liquid argon time projectionchamber (LArTPC) experiments helps people exchange ideas and test algorithms on realdata. This work has been focused on single-phase TPCs. The progress on implementingdual-phase TPC simulation and reconstruction in LArSoft will be described in Section 3.
2.1 Monte Carlo Challenge
We produce large samples of Monte Carlo events (Monte Carlo Challenge - MCC) on a regularbasis. The purpose is to test the latest simulation and reconstruction chain as well as thegrid job submission tools and sam/tape interfaces and provide standard samples for variousphysics working groups to analyze. All the MCC samples are produced using the batch toollarbatch [4]. MCC1.0 was produced in January 2015 with three 35t samples. We includefar detector and protoDUNE samples in later MCCs and incorporate more sophisticateddetector simulation and reconstruction software. The latest MC production MCC6.0, whichwas produced in May 2016, consists of 50 samples using the 35t, far detector and protoDUNEsingle-phase geometries. MCC6.0 was produced using LArSoft v05 09 01. The locations ofthe samples can be found in Ref [5].
In MCC6.0, we simulated 3 types of beam neutrinos (unoscillated νµ’s, fully-oscillatedνe’s and fully-oscillated ντ ’s), atmospheric neutrinos, supernova neutrinos, proton decays andcosmogenic events in the far detector. In order to save processing time, all the far detectorsamples except the cosmogenics sample were simulated using a smaller version of the full 10kt far detector geometry. This geometry is 13.9 m long, 12 m high and 13.3 m wide, whichconsists of 12 APAs and 24 TPCs. In the default configuration, the TPC wire spacing is5 mm, the wire angle is 36◦ and the neutrino beam is parallel to the wire planes. We alsogenerated special samples where the wire spacing is 3 mm or the wire angle is 45◦ or theneutrino beam is perpendicular to the wire planes for the detector optimization studies.
3
2.2 Simulation Chain
Each sample is simulated in 3 steps: generation (gen), geant4 tracking (g4) and digitization(detsim). The first step is unique for each sample while the second and the third steps arecommon for all samples.
2.2.1 Generation
Both the beam neutrinos and atmospheric neutrinos were simulated using genie v2 10 6.The beam neutrinos were simulated using the reference flux. We simulated unoscillatedsamples in both the neutrino and antineutrino modes. We also simulated fully oscillatedsamples where we convert all νµ’s in the beam to either νe’s or ντ ’s.
The atmospheric neutrinos were simulated using the Bartol flux for the Soudan site [6].There is a plan to use the Honda flux for the Homestake site [7] when the genie flux interfaceis updated.
The proton decay events were simulated using standalone genie v2 10 6 after a bug inthe kaon final state interaction simulation was identified and fixed. Only the decay modep → ν̄ + K+ was simulated. The output files were converted to the LArSoft format forfurther processing.
The supernova neutrino events were generated using custom code wrapped in a LArSoftmodule (also described in Section 4.1.3). This code simulates charged-current νe-
40Ar in-teractions. For each electron neutrino it calculates probabilities to produce a 40K nucleusin different excited states (using a model from [8]), randomly selects one, and (with energylevels from [9]) produces several de-excitation γs and an electron carrying the remainingenergy. All particles are produced isotropically, there is no delay between the electron andcorresponding de-excitation γs (in this model the 40K nucleus de-excites instantaneously)and they share a vertex, which is simulated with equal probability anywhere in the activevolume. The primary neutrino energy distribution used in these samples is the cross-section-weighted energy spectrum obtained from SNOwGLoBES [10] (using the “GKVM” flux [11]).The supernova neutrino generator also allows to simulate a Poisson-distributed random num-ber of neutrino interactions per event. These samples were simulated with, on average, 2or 20 neutrinos. In addition, one of the samples was generated with 1.01 Bq/kg of 39Arbackground.
Cosmogenic events at depth were generated using MUSIC (Muon Simulation Code) [12]and MUSUN (Muon Simulations Underground) [13]. MUSIC first propagates a set of muonsthrough a medium with user-specified parameters (e.g. density, elemental composition),given initial parameters of energy, position and direction cosines. It takes muons of ener-gies ranging from 102 to 107 GeV and stores their energy distribution at between 100 and15000 meters-water-equivalent (mwe). MUSUN then generates a muon energy spectrum andangular distribution appropriate for an underground location using the output of MUSIC aswell as details about the local surface profile. This information is used to calculate the totalamount of rock a muon would traverse through given its inclination. The particles passed togeant4 are then sampled on the surface of a box surrounding the detector and containinga few meters of rock. The average energy of the muons simulated is 284 GeV.
4
2.2.2 geant4 Tracking
The truth particles generated in the event generator step are passed to a geant4 v4 10 1 p03based detector simulation. In this step, each primary particle from the generator and its de-cay or interaction daughter particles are tracked when they traverse liquid argon. The energydeposition is converted to ionization electrons and scintillation photons. Some electrons arerecombined with the positive ions [14, 15] while the rest of the electrons are drifted towardsthe wire planes. The number of electrons is further reduced due to the existence of impuritiesin the liquid argon, which is commonly parameterized as the electron lifetime. The defaultelectron lifetime is 3 ms in the simulation. The longitudinal diffusion smears the arrival timeof the electrons at the wires and the transverse diffusion smears the electron location amongneighboring wires. More details regarding the recent measurements of diffusion coefficientscan be found in Ref. [16, 17].
When ionization is calculated, the amount of scintillation light is also calculated. Theresponse of the photon detectors is simulated using a ‘photon library,’ a pre-generated tablegiving the likelihood that photons produced within a voxel in the detector volume will reachany of the photon detectors. The photon library is generated using Geant4’s photon trans-port simulation, including 66 cm scattering length, 20 m attenuation length, and reflectionsoff of the interior surface detectors. The library also incorporates the response vs. locationof the photon detectors, capturing the attenuation between the initial conversion location ofthe photon and the SiPMs.
2.2.3 Digitization
The electrons on each wire are converted into raw wire signal (ADC vs Time) by convolutionwith the field response and electronics response. The field response on each wire plane issimulated with Garfield [18] while the ASIC electronics response was simulated with theBNL SPICE [19] simulation. Currently, the signal on each wire can only be produced fromthe ionization electrons going through the wire. The improvement upon this approximationis ongoing. This in particular is important for the induction wire planes, where the inductionsignal depends on the local ionization charge distribution [20]. For most samples, the ASICgain was set to 14 mV/fC and the shaping time was set to 2 µs. For the samples generatedwith 3 mm wire spacing, the ASIC gain was set to 25 mV/fC. The noise level was set to2.5 ADC RMS. In the current simulation, the electronic noise is assumed to be white, whichis a uniform distribution in the frequency domain. The implementation of a more realisticelectronics noise model is ongoing. Figures 1 and 2 show the expected electronics shapingfunctions and simulated field response for the closest wire, respectively.
The photon detector electronics simulation separately generates waveforms for each chan-nel (SiPM) of a photon detector that has been hit by photons. Every photon that has beendetected appears as a single photoelectron pulse (with the shape taken from [21]) on arandomly selected channel (belonging to the photon detector in which the photon was reg-istered). Then dark noise (with the rate of 10 Hz) and line noise (Gaussian noise with theRMS of 2.6 ADC counts) are added. Each photon (or a dark-noise pulse) has a probability ofappearing as 2 photoelectrons on a waveform (the cross-talk probability is 16.5 %). The finalstep of the digitization process is recording only fragments of the full simulated waveforms
5
Figure 1: ASIC’s electronics shaping functions are shown for four shaping time settings at14 mV/fC gain.
Figure 2: Field response simulated with Garfield program is shown for two induction andone collection planes.
6
that have a signal in them. That is accomplished by passing the waveforms through a hitfinder described in Section 2.3.11 and storing parts of the waveforms corresponding to thehits found.
2.3 Reconstruction Chain
In this section, we describe various reconstruction algorithms used to reconstruct events inthe far detector TPC.
2.3.1 Signal Processing
The raw data are in the format of ADC counts as a function of TPC ticks on each wire. Thesignal can have a unipolar shape if it is on a collection wire or a bipolar shape if it is on aninduction wire. The first step in the reconstruction is to convert the raw signal from eachwire to a standard shape such as a Gaussian shape. This is achieved by passing the raw datathrough a calibrated deconvolution algorithm. In real detectors, excess noises may exist andneed to be properly dealt with through a dedicated noise filter [22]. There is currently nosuch complication in the simulation.
The deconvolution technique was introduced to LArTPC signal processing by BruceBaller in the context of ArgoNeuT data analysis. The goal of the deconvolution [23] is to“remove” the impact of field and electronics responses from the measured signal to recoverthe number of ionized electrons. This technique has the advantages of being robust and fastand is an essential step in the overall drifted-charge profiling process.
Deconvolution is a mathematical technique to extract a real signal S(t) from a measuredsignal M(t0). The measured signal is modeled as a convolution integral over the real signalS(t) and a given detector response function R(t, t0) which gives the instantaneous portionof the measured signal at some time t0 due to an element of real signal at time t:
M(t0) =
∫ ∞−∞
R(t, t0) · S(t) · dt. (1)
If the detector response function only depends on the relative time difference between t andt0,
R(t, t0) ≡ R(t− t0), (2)
we can solve the above equation by doing a Fourier transformation on both sides of theequation:
M(ω) = R(ω) · S(ω), (3)
where ω is the frequency. In this case, we can derive the signal in the frequency domain bytaking the ratio of measured signal and the given response function:
S(ω) =M(ω)
R(ω). (4)
The real signal in the time domain can then be obtained by applying the inverse Fouriertransformation from the frequency domain.
7
The response function R(ω) does not address contributions to the measured signal whichare due to real world sources of electrical noise from thermal and unwanted transmittingsources or the approximation in the digitization. Such contributions to M(ω) will not beremoved by the deconvolution. Worse, because the response function becomes small atlow frequencies for the induction planes and at high frequencies for all planes, the noisecomponents in these frequencies will be enhanced by the deconvolution.
To address the issue of noise, a filter function F (ω) is introduced. Its purpose is to atten-uate the problematic noise. The addition of this function can be considered an augmentationto the response function. The two functions are kept distinct for clarity in the notation here.Equation (4) is then updated to become
S(ω) =M(ω)
R(ω)· F (ω). (5)
With a suitable noise filtering model an improved estimator for the signal S(t) in the timedomain can then be found by applying an inverse Fourier transform to S(ω). Essentially,the deconvolution replaces the real field and electronics response function with an effectivesoftware filter response function. The advantage of this procedure is more pronounced on theinduction plane where the irregular bipolar field response function is replaced by a regularuni-polar response function through the inclusion of the software filter.
Since the current simulation only simulates induced current on the closest wire to theionization electrons, the aforementioned 1-D deconvolution technique is sufficient to processthe TPC signal. However, the induction signal in a real detector is expected to stronglydepend on the local ionization charge distribution. In this case, the 1-D deconvolution isnot enough to recover the number of ionization electrons. A 2-D deconvolution technique,which includes both the time and wire dimensions, has been developed in MicroBooNE todeal with the complicated induction plane signal. Details can be found in Ref. [20].
2.3.2 Gaussian Hit Finder
GausHitFinder is a hit-finding algorithm that works starting from deconvolved signals onwires and defining areas above threshold as “pulses”. Once a pulse is found, an ‘n’ Gaussianhypothesis is applied where ‘n’ is defined by the number of peaks initially identified withinthe pulse. Based on the outcome of the fit an object known as a “hit” is formed and storedin the event.
2.3.3 Disambiguation
The induction plane wires are wrapped in the DUNE TPC design in order to save cost onelectronics and minimize dead region between APAs. The consequence is that multiple in-duction wire segments will be read out by the same electronic channel. We need to determinewhich wire segment corresponds to the energy deposited by the particle in the TPC. Thisprocess is called disambiguation.
The current disambiguation algorithm was developed for the 35t geometry. It relies onthe fact that the collection plane wires are not wrapped and the wire angles are slightlydifferent for the two inductions views (44.3◦ vs 45.7◦) so that any three wires from the three
8
planes read out by the same three channels will never cross twice. The algorithm uses gaushitas input. All the collection plane hits are unambiguous. Each induction plane hit has onechannel ID and several possible wire IDs corresponding to various wire segments read out bythe same channel. We need to determine which wire ID is the correct one for each inductionplane hit.
The disambiguation algorithm first loops over all collection plane hits. For each collectionplane hit, it loops over all induction plane hits and looks for one hit on each of the twoinduction planes that are in time with the collection plane hit. Once the triplet of hits isfound (one on each plane with a common time), the algorithm checks all possible wire IDsand looks for three wires that intersect. Once one and only one intersection is identified, thetwo induction wire IDs are assigned to the two induction plane hits and the ambiguity isresolved. Finally the algorithm loops over all unresolved induction plane hits. For each hit,it loops over all possible wire segments for that channel and chooses the wire segment thatis closest to a resolved induction hit as the correct wire segment.
The same disambiguation algorithm also works for the far detector and protoDUNE.
2.3.4 Line Cluster
The intent of the Line Cluster algorithm is to construct two-dimensional line-like clustersusing local information. The algorithm was originally known as Cluster Crawler. The“Crawler” name is derived from the similarity of this technique to “gliders” in 2D cellularautomata. The concept is to construct a short line-like “seed” cluster of proximate hits inan area of low hit density where hit proximity is a good indication that the hits are indeedassociated with each other. Additional nearby hits are attached to the leading edge of thecluster if they are similar to the hits already attached to it. The conditions are that theimpact parameter between a prospective hit and the cluster projection is similar to thosepreviously added and the hit charge is similar as well. These conditions are moderated toinclude high charge hits that are produced by large dE/dx fluctuations and the rapid increasein dE/dx at the end of stopping tracks while rejecting large charge hits from δ-rays. Seedclusters are formed at one end of the hit collection so that crawling in only one directionis sufficient. Line Cluster uses disambiguated gaushits as input and produces a new set ofrefined hits. More details on the Line Cluster algorithm can be found in Ref [24].
2.3.5 Blurred Cluster
The Blurred Cluster reconstruction method aims to construct two-dimensional shower-likeclusters from deposits left in the detector by showers. It specializes in shower reconstruction,especially in the separation of nearby showers in the reconstruction of, e.g., π0 decay. Thealgorithm first applies a weighted 2D Gaussian smearing to the hit map in order to introduce‘fake hits’ and distribute the charge deposited in the detector more realistically. This pro-ceeds by convolution of a Gaussian kernel, uniquely applied for each hit given informationsuch as rough directionality of the showering particles and the width of the reconstructedhits in time in order to introduce the most accurate blurring possible. Clustering followsby grouping neighboring hits within the blurred region before removing any artificial hitsand forming output clusters from the remaining hit collections. BlurredCluster uses disam-
9
biguated gaushit as input and the output clusters are in turn used as input to the EMShoweralgorithm (see Sec. 2.3.8). More details and discussion are available in Ref [25].
2.3.6 Pandora
The Pandora Software Development Kit [26] was created to address the problem of identifyingenergy deposits from individual particles in fine-granularity detectors. It promotes the ideaof a multi-algorithm approach to solving pattern-recognition problems. In this approach,the input building-blocks (Hits) describing the pattern-recognition problem are consideredby large numbers of decoupled algorithms. Each algorithm targets a specific event topologyand controls operations such as collecting Hits together in Clusters, merging or splittingClusters, or collecting Clusters in order to build Particles. The output from the chain ofover 70 algorithms is a hierarchy of reconstructed 3D Particles, each with an identifiedparticle type, vertex and direction.
2.3.7 Projection Matching Algorithm
Projection Matching Algorithm (PMA) was primarily developed as a technique of 3D recon-struction of individual particle trajectories (trajectory fit) Ref [27]. PMA was designed toaddress a challenging issue of transformation from a set of independently reconstructed 2Dprojections of objects into a 3D representation. Reconstructed 3D objects are also providingbasic physics quantities like particle directions and dE/dx evolution along the trajectories.PMA uses as its input the output from 2D pattern recognition: clusters of hits. For the pur-poses of the DUNE reconstruction chain the Line Cluster algorithm (2.3.4) is used as inputto PMA; however the use of hit clusters prepared with other algorithms may be configuredas well. As a result of 2D pattern recognition, particles may be broken in 2D projections intoseveral clusters, fractions of particles may be missing in individual projections and clustersobtained from complementary projections are not guaranteed to cover corresponding sectionsof trajectories. Such behavior is expected since ambiguous configurations of trajectories canbe resolved only if the information from multiple 2D projections is used. Searching for thebest matching combinations of clusters from all 2D projections was introduced with PMAimplementation in the LArSoft framework. The algorithm attempts also to correct assign-ments of hits to clusters using properties of 3D reconstructed objects. In this sense PMA isalso a pattern recognition algorithm.
The PMA underlying idea is to build and optimize objects in 3D space (formed as polyg-onal lines with iteratively increased number of segments) by minimizing the cost functioncalculated simultaneously in all available 2D projections. The cost function consists of 2Ddistance of hits to the optimized object 2D projections, penalty of tracks curvature, and 3Ddistance of various feature points to the optimized object (used e.g. to improve performancefor tracks with isochronous orientation). The track can be reconstructed using clusters fromtwo projections while the distance of hits to the track projection in the third plane is usedto validate correct association of clusters. This method is used to score 3D track candidatesin the three-plane TPC configurations, like single-phase DUNE and MicroBooNE detectorsand prototypes. Clusters from all planes are used in this scenario in the fine tuning of the se-lected candidates. In the two-plane configurations (dual-phase TPC detectors, single-phase
10
LArIAT and ArgoNeut detectors) the track candidates are scored by the value of cost func-tion, which is significantly increased if the trajectory fit is being optimized to spuriouslyassociated clusters (this is also a second-level criterion for the three-plane configurations ifthe validation method shows no significant difference for track candidates).
The approach of constructing entire objects in 3D allows us to avoid the requirement offinding associations between 2D planes on the level of individual hits. Such associations areespecially problematic if several hits on a single trajectory can be found with a similar drifttime value, or, if due to a low signal (or any other reason) hits are missed in one of projections.Unambiguous 3D position is calculated for each 2D hit, independently from other hits in thetrajectory. Such 3D positions are more accurate than found with the “standard” calculationof a single 3D point using multiple 2D hits matched by drift time values, leading to the moreaccurate dE/dx estimation. It also allows us to build 3D objects using the detector datafrom 2D planes directly, without an intermediate step of 3D points calculation which aresubsequently used to obtain the final trajectory fit. Another advantage of the approach isthe capability of estimating the direction of small, few-hit tracks, or estimating the 3D PCaxis of shower-like objects.
Several features were developed in LArSoft’s PMA implementation in order to addressdetector-specific issues like stitching the particle fragments found in different TPC’s or anoption for performing disambiguation at the 3D reconstruction stage. Since algorithms ex-isting within the LArSoft framework or interfaced to it (see Sec. 2.3.6) can provide patternreconstruction results which include the particle hierarchy description, the mode of apply-ing PMA in order to calculate solely the trajectory fits was developed. In this mode thecollections of clusters forming particles are taken from the “upstream” algorithm and alsoassociations of hits to clusters are not changed.
Recent developments of PMA rely on the same principal ideas and allow us to build andoptimize complex structures of 3D objects, i.e. multiple particle trajectories interconnectedwith interaction vertices. With such an approach it is possible to employ in the vertexposition reconstruction the local information from several tracks simultaneously, leading alsoto an improved fit of each individual trajectory. The track-vertex structure is constructedafter the individual trajectories are found and stitched across TPCs. Vertex candidates arefound as regions of intersection of two or more tracks (including regions found beyond thetrack endpoints), with the threshold on the allowed region size. Tracks shorter than nn cmare treated separately; they can be associated to vertex candidates found using long tracksin the first pass of the algorithm or used to create vertex candidates in the second pass.Candidates are scored by the number of intersecting tracks, and if this number is equal fortwo candidates the maximum angle between intersecting tracks is used to select candidatesand create a better defined vertex. The entire track-vertex structure with the newly createdvertex is re-optimized using PMA principles in order to accommodate additional informationin the trajectory fits. Since the new vertices may connect structures that already containvertices, a set of rules for splitting and flipping tracks was developed to ensure that theresulting structure is always tree-like (i.e. does not contain loops). The resulting structureis described as a particle hierarchy using data products available in LArSoft.
Two important reconstruction features are currently under development in LArSoft’sPMA implementation. The first is the integration of PMA input with the recognitionof shower-like 2D objects (electromagnetic showers, electron tracks) and track-like objects
11
(hadron and muon tracks). Such recognition is a prerequisite of obtaining robust 3D trackingwith PMA, since different fitting strategies should be applied to track and shower objects.The second feature under development is the detection of kinks and decay points missed atthe 2D pattern recognition level. First attempts were made with the search for outliers inthe distribution of angles in the trajectory fit. It was found that the dependence on the trackorientation in 2D projection needs to be taken into account; also a potential integration withan algorithm based on 2D ADC image analysis will be explored.
2.3.8 EMShower
The EMShower reconstruction algorithm aims to find final 3D showers and all associatedproperties. It is intentionally high-level by design and relies heavily on previous reconstruc-tion, specifically BlurredCluster (Sec. 2.3.5) and PMA (Sec. 2.3.7). The reconstructionproceeds in two general steps: first, the shower objects, including all associated hits in eachof the views, are found; second, the properties of these showers, such as start point, direction,energy and initial dE/dx, are determined by multiple pattern recognition and calorimetricreconstruction algorithms.
The shower objects are created by simply matching the previously found, well-formed,shower-like clusters (provided by BlurredCluster) between the different views to form 3Dobjects with associated hits in each plane. This is achieved by associating hits betweenthe 2D shower-like clusters and 3D tracks (provided by, e.g., PMA) in order to pull togetherclusters from different planes into one object. These shower hits are then analyzed by varioussuccessive algorithms in order to find relevant properties before the output shower objectsare constructed for later use.
Further details and discussion can be found in Ref [28].
2.3.9 Calorimetric Energy Reconstruction and Particle Identification
As charged particles traverse a liquid argon volume, they deposit energy through ionizationand scintillation. It is important to measure the energy deposition as it provides informationon particle energy and species. The algorithm for reconstructing the ionization energy inLArSoft is optimized for line-like tracks and is being extended to more complicated eventtopology such as showers. The algorithm takes all the hits associated with a reconstructedtrack. For each hit, the hit area or amplitude, in ADC counts, is converted to the chargeQdet, in units of fC, on the wire using an ADC to fC conversion factor that was determinedby muons or test stand measurements. To account for the charge loss along the drift dueto impurities, a first correction is applied to Qdet to get the free charge after recombinationQfree = Qdet/e
−t/τe , where t is the electron drift time for the hit and τe is the electronlifetime measured by the muons or purity monitors. The charge Qfree is divided by the trackpitch dx, which is defined as wire spacing divided by the cosine of the angle between thetrack direction and the direction normal to the wire direction in the wire plane, to get thedQfree/dx for the hit. Finally, to account for charge loss due to recombination, also knownas “charge quenching”, a second correction is applied to convert dQfree/dx to dE/dx basedon the modified Box’s model [14] or the Birks’s model[15]. The total energy deposition from
12
the track is obtained by summing the dE/dx from each hit:all hits∑
i
(dE/dx)i · dxi.
If the incident particle stops in the LArTPC active volume, the energy loss, dE/dx,as a function of the residual range (R), the path length to the end point of the track, isused as a powerful method for particle identification. There are two methods in LArSoftto determine particle species using calorimetric information. The first method calculatesfour χ2 values for each track by comparing measured dE/dx versus R points to the proton,charged kaon, charged pion and muon hypotheses and identifies the track as the particlethat gives the smallest χ2 value. The second method calculates the quantity PIDA =<Ai >=< (dE/dx)iR
0.42i > [14], which is defined to be the average of Ai = (dE/dx)iR
0.42i over
all track points where the residual range Ri is less than 30 cm. The particle species can bedetermined by making a selection on the PIDA value.
2.3.10 WireCell
WireCell [29], which adopts a very different approach from the aforementioned algorithms, isa new reconstruction method under development. Instead of directly doing pattern recogni-tion on each of the 2D views (drift time vs. wire number), the first step of the WireCell recon-struction is to perform 3D imaging with time, geometry, and charge information. The defini-tion of ”Hit” is based on signal strength after charge extraction in a 2 µs time slice. The usageof time information means that hits from different wire planes at different time slices cannotbe associated. The usage of geometry information means that hits from wires that are notcrossing each other cannot be associated together. The usage of charge information meansthat the hits from different wire planes with different signal strengths are unlikely to be asso-ciated together. The usage of the charge information is quite unique for a LArTPC, as eachof the wire planes in principle detects the same amount of the ionization electrons. Figure 3shows the performance of the WireCell 3D imaging. The large blue blob reflects the ambigu-ities due to wire readout for a track traveling parallel to the wire plane. (The 3D web-baseddisplay can be found at http://www.phy.bnl.gov/wire-cell/bee/set/6/event/20/.) Inthis case, a group of wires from each wire plane is fired simultaneously. Therefore, the timeand geometry information will provide rather limited constraints on the hit associations.The advantage of the WireCell approach is that it utilizes full TPC information. The strongrequirement of the time/geometry/charge information provides a natural way to suppresselectronic noise at the cost of being more sensitive to the hit inefficiency. Since the trackand shower hypotheses are not used, the 3D imaging works for any event topology. Once the3D images are reconstructed, 3D pattern recognition is needed to identify the content insidethe image. Figure 4 shows the performance of the currently available 3D pattern recogni-tion inside WireCell. More developments on the WireCell pattern recognition are neededbefore calculation of physics quantities will be possible. At the moment, the development ofWireCell relies on the improvement in the TPC signal processing as well as the 3D patternrecognition techniques.
In the following, we briefly describe the concept of the WireCell tomographic imagingstep. Calibrated WireCell imaging determines the likely spatial distribution of activity in thedetector volume which is consistent with the measured signals. The distribution is definedon a collection of voxels filling the volume. The voxels are shaped as polygonal, right-angled
13
Figure 3: Comparison of imaging reconstruction qualities with and without the charge in-formation.
14
Figure 4: The reconstructed image is shown on the left panel for one neutrino interactionevent. The image was passed through the 3D pattern recognition program with tracksidentified (middle panel). The identified pattern is compared with Monte-Carlo truth (rightpanel).
extrusions. The two polygonal faces are called “cells” and are parallel to the wire planes.A cell covers the region near a triple-overlap of one wire from each plane. Imagine a strip
surrounding each wire with width ±12
the wire pitch which runs parallel to the wire and inthe wire plane. A cell is formed by the intersection of three such strips, one from a wire ineach plane.
The sides of the voxels are parallel to the drift and extend to cover the distance electronswill drift during one “time slice” (chosen to be four time ticks due to expected timingresolution.).
The measured signal on a wire is subject to a threshold. For a given time slice, if thesignal is above this threshold the wire is considered “hit”. Likewise, for each time slice, anycell which has its three associated wires “hit” is itself considered “hit”. 1 Given the timeof the hit cell and external information about the initial interaction time (from the opticaldetector system), the cell can be projected into the detector volume along the drift path toprovide the location of the voxel.
One of the unique features of the LArTPC is that each wire plane, be it induction orcollection, provides a measure of the same distribution of ionization electrons. This featureis independent of the initial event topology (i.e. track angle, shower, vertex etc.). Therefore,this feature can be used to remove ambiguities made by using wire planes for the readout.Using the concept of wire and cell, we can thus write down the following charge equation:
B ·W = G · C. (6)
1In the real world case where some channels do not read out, the signal requirement is relaxed to considerthe cell hit if at least two wires are hit and the third wire is a known dead channel. This leads to complicationssuch as producing artificial cell hits (“ghosting”).
15
Here, W represents a vector of charge measured in wires from all wire planes in a given timeslice. B is a matrix connecting merged wires with single wires 2. C is a vector representingthe likely amount of ionization electrons in merged cells or blobs 3. G is the geometrymatrix connecting merged cells and the merged wires. This equation can be expanded intoa chi-square function:
χ2 = (B ·W −G · C)T V −1BW (B ·W −G · C) , (7)
which also takes into account the uncertainties of the measured charge in wires. In particular,VBW ≡ B · VW · BT is the covariance matrix describing the uncertainty in (merged) wirecharges.
The minimum of the above chi-square function can be found by calculating the firstderivative
∂χ2
∂C= 0→ GTV −1BW (B ·W −G · C) + (B ·W −G · C)T V −1BWG = 0, (8)
and the solution can be written as:
C =(GT · V −1BW ·G
)−1 ·GT · V −1BW ·B ·W. (9)
The core of the Eq. (9) is the inversion of the matrix GT · V −1BW · G. When this matrix canbe inverted, the charge of merged cells can be derived directly. For faked hits (merged cellswithout any ionization charge), the derived charge is likely to be close to zero. For real hits(merged cells with ionization charge), the derived charge is likely to be large and close tothe actual true value. On the other hand, if this matrix can not be inverted, additionalassumptions and more advanced techniques are needed to derive the solution. The detailsof these techniques are beyond the scope of this report, and will not be discussed here.
2.3.11 Optical Reconstruction
Optical Hit Finder
The first step of the DUNE optical reconstruction is reading individual waveforms fromthe simulated photon-detector electronics and finding optical hits – regions of the waveformscontaining pulses. The optical hit contains the optical channel (SiPM) that the hit was foundon, time corresponding to the hit, its width, area, amplitude, and number of photoelectrons.
The current DUNE optical-hit-finder algorithm searches for regions of the waveform ex-ceeding a certain threshold (13 ADC counts), checking whether that region is wider than10 optical TDC ticks, and, if it is, calculating the aforementioned optical-hit parameters forthe region (including parts of the waveform around it that have ADC values greater than 1)and recording it as an optical hit. The number of photoelectrons is calculated by dividing
2A group of “merged wires” is formed by collecting together all neighboring hit wires in the time slice.This is done to reduce the rank of the matrix equation.
3“A group of “merged cells” is formed by collecting together all primitive cells which are hit and self-contiguous.
16
the full area of the hit by the area of a single-photoelectron pulse. The pedestal is assumedto be constant and is specified in the hit finder as 1500 ADC counts.
Optical Flash Finder
After optical hits are reconstructed, they are grouped into higher-level objects called op-tical flashes. The optical flash contains the time and time width of the flash, its approximateY and Z coordinates (and spatial widths along those axes), its location and size in the wireplanes, the distribution of photoelectrons across all photon detectors, and the total numberof photoelectrons in the flash, among other parameters.
The flash-finding algorithm searches for an increase in photon-detector activity (the num-ber of photoelectrons) in time using information from optical hits on all photon detectors.When a collection of hits with the total number of photoelectrons greater than or equal to2 is found, the algorithm begins creating an optical flash. It starts with the largest hit andadds hits from the found hit collection that lie closer than 0.5 of the combined widths of theflash under construction and the hit being added to it. The flash is stored after no more hitscan be added to it and if it has more than 2 photoelectrons.
The algorithm also estimates spatial parameters of the optical flash by calculating thenumber-of-photoelectron-weighted mean and root mean square of locations of the opticalhits (defined as centers of photon detectors where those hits were detected) contained in theflash.
3 Dual Phase Simulation and Reconstruction Chain
Up until recently, LArSoft had only been used for simulation of single-phase TPCs. Softwarefor dual-phase TPCs existed for the WA105 experiment, but had not been incorporated intoLArSoft. This section describes the implementation of dual-phase simulation and recon-struction into the LArSoft framework for DUNE.
The design of the 10kt dual-phase (DP) detector module envisions a LArTPC with afully active volume of 12 × 12 × 60 m3. Ionization charges produced in the active volumeare drifted vertically upwards toward the liquid-gas boundary, extracted into the gaseousphase, amplified in the large electron multiplier (LEM), and collected on the anode with twoorthogonal collection views having a pitch of 3.125 mm. The charge readout over the 12 ×60 m2 area is accomplished with independent 3× 3 m2 modules containing the sandwichedLEM-anode tiles of 50× 50 cm2 and the grid for the charge extraction.
To model this segmentation of the charge readout plane (CRP) in the LArSoft simulation,one introduces independent TPC units with 12 m drift and readout area of 3 × 3 m2 each.Fig.5 illustrates the implemented geometry in the case of 10 kt DP module. In the simulationthe x -axis points in the direction of the electron drift, the y-axis is along the vertical, andz -axis is in the neutrino beam direction. Therefore, the charge readout is in xz -plane. Itshould be noted that while this orientation of the coordinate axes is unphysical (i.e., the Argas layer appears on the side of the detector instead of the top), it is necessary in order tomake geometry compatible with:
• LArSoft requirement that the drift field is along ±x,
17
• Definition of axis orientation already in use for geometry description of the single-phaseTPC.
The origin of the coordinate system is set to be in the middle of the upstream face of theactive volume similar to the geometry description of the single-phase detector option.
Figure 5: Illustration of the basic geometry for the 10kt DP TPC implemented in LArSoft.The active volume is in green and surrounding LAr buffer in light grey. The drift directionis along x, while the neutrino beam is in the z direction.
The overall dimensions of each CRP module in the simulation are set by the active areaof each unit (3 × 3 m2) plus the border. The latter is taken to be 5 mm based on therequirement for the CRP mechanical design. Such border width then results in a gap of10 mm space between adjacent modules.
To simulate charge readout in LArSoft, the framework requires a consecutive collection ofwire planes. This approach is well-suited for the description of single-phase detector, wherethe electron cloud is drifted in liquid argon across the induction views to the collection wires,but does not correspond to the case of the dual-phase detector, where the segmented anodecollects charge simultaneously in two orthogonal views in the gaseous phase. To meet theframework requirements, however, the DP anode is modeled as a two thin ”wire planes”0.3 mm thick. A dummy electric field of 4.0 kV/cm (vdrift ∼ 3.5 mm/µs in LAr), themaximal value for which the drift velocity calculation is valid in LArSoft, is applied betweenthe two wire planes in order to allow drift of electrons from one collection plane to the other.This in effect introduces a small time difference between two collection views given by theelectron drift time in the first wire plane:
∆t = tview1 − tview0 =0.3 mm
3.5 mm/µs= 86 ns.
However, the offset is small compared 2.5 MHz ADC sampling rate (400 ns time bin) used by
18
the DP electronics and at most would correspond to 1 time tick difference in charge arrivaltimes.
The charge amplification performed by the LEM is accomplished in the simulation byscaling true charge recorded on a given readout channel at a given time after taking intoaccount electron lifetime and diffusion effects by the factor (DPhaseSimChannelExtractSer-vice.DPGainPerView) that parametrizes the expected gain per each collection view. Aftermultiplying the collected charge by the gain factor the result is convoluted with the front-endamplifier response function, noise and pedestals are added, and the resultant signals are thenquantized to generate expected raw ADC waveforms for each channel.
The basic checks of spatial and calorimetric reconstruction have been performed withthe current implementation of dual-phase far detector geometry and detector simulationin LArSoft. The aim was to apply and validate the existing hit finding and hit fittingalgorithms in LArSoft as a first step as these provide mandatory input to the higher levelpattern recognition and 3D reconstruction (spatial and calorimetric).
Tests were performed using samples of 2 GeV/c muons simulated at different depths inthe detector (drift distances), using the default value for the electron lifetime value of 3 ms.The aim was to reconstruct m.i.p. scale from relatively straight tracks that generate signalsof different amplitudes and widths due to the charge attenuation by LAr impurities anddiffusion of drifting electrons in LAr.
It was observed that the raw ADC waveforms of wire signals obtained with the defaultsimulation parameters for the electron drift contained high fluctuations. This behavior wasrelated to the procedure of diffusion simulation, where the spatial position and drift time ofionization electrons projected to wire planes is calculated for bunches (clusters) of electrons.The default number of electrons per cluster was too high for large drift distances givinga small number of clusters and thus resulting in a poor approximation of diffusion effects.Simply decreasing the number of electrons per cluster solves the issue but may lead toincrease of computation time in case of busy events. Therefore a parameterized limit on theminimum number of clusters has been added to the LArSoft simulation code.
The standard hit reconstruction implemented in LArSoft consists of two steps:
• Deconvolution of the E-field and electronics response,
• Hit finding and fitting with the assumption of Gaussian shape of ionization signals.
While the deconvolution may include additional filters for the noise reduction, the implemen-tation of deconvolution procedure itself acts as a sharp low-pass filter by removing frequenciesat which electronics response or E-field response have too low an amplitude in the Fourierspectrum. The rationale is to avoid division by small values, which are likely estimated withhigh uncertainty. In practice such cuts lead to the removal of high frequency componentsleading to ringing oscillation artifacts around deconvolved signal pulses. Amplitudes of suchoscillations at small drift distance are comparable to the signal peak amplitude at long driftdistance making it difficult to reject these artifacts with a simple threshold cut during hit re-construction. These oscillations can, however, be minimized with an additional filter. Fig. 6shows the frequency spectrum of the electronics impulse response overlaid with the spectrumof the filter function adapted for the DP detector.
19
MHz0.2 0.4 0.6 0.8 1
Amplitude
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
Figure 6: Fourier transform of the impulse response from DP electronics (blue) with filterresponse (red) overlaid.
One issue stemming from the deconvolution is to relate the estimated area under a re-constructed hit to the number of electrons. In principle, this calibration constant (Calorime-tryAlg.CalAreaConstants) should be completely set by the readout chain (LEM gain, cali-brated front-end electronics amplifier gain and response function). However, in the case ofdeconvolution, the required filtering leads to cuts in the frequency domain thus resultingin information loss. One, therefore, finds this calibration factor by tuning its value untilthe peak of the reconstructed dE/dx matches the m.i.p. expectation. Fig. 7 shows the re-constructed dE/dx for 2 GeV/c muons after the calibration constant is tuned to the m.i.p.scale.
Entrieso 1723219
Meano 2.226
dE/dxo[MeV/cm]0 2 4 6 8 10
0
50
100
150
200
250
300310×
Entrieso 1723219
Meano 2.226
muonsocloseotoocathode
Entrieso 1524788
Meano 2.349
dE/dxo[MeV/cm]0 2 4 6 8 10
0
50
100
150
200
250
310×
Entrieso 1524788
Meano 2.349
muonsocloseotooCRM
Figure 7: Reconstructed dE/dx from 2 GeV/c muons with electron lifetime of 3 ms. Left:tracks near to the cathode (large charge loss). Right: tracks near charge readout plane (smallcharge loss).
The use of the deconvolution for hit reconstruction is strongly motivated by bi-polar
20
nature in the response of induction planes in single-phase LAr TPC detectors (in general thebi-polar inductive response is also present in the collection view due to the signal inducedon the neighboring strips/wires around a given collection channel; however, the effect isnegligible). For the collection view an approach that relies on hit finding and fitting inthe time domain (raw waveforms without deconvolution) is also suitable and could be usedfor the dual-phase detector since both views are collection views. The fit function in thiscase should include the signal shaping due to electronics response. An implementation ofsuch algorithm has been proposed to be added to LArSoft hit reconstruction tool-kit and iscurrently under discussion.
Once the hit finding parameters are appropriately tuned the complete chain from thesimulation up to 3D reconstruction can be run successfully with the dual-phase geometry.The reconstruction algorithms were tested on a sample of stopping muons (0.2 GeV/c) andprotons (2 GeV/c). These samples contain several classes of low complexity topologies. Suchtopologies are expected to be reconstructed in detail, allowing analysis of interaction verticesand providing dE/dx evolution along the particle trajectories. The following reconstructionchain has been tested with the dual-phase detector simulation: Gauss hit fitting, lineclusterserving as 2D pattern recognition to find objects in both planes, and Projection MatchingAlgorithm serving as 3D pattern recognition together with 3D trajectory fitting and vertexreconstruction. Fig. 8 shows the example of reconstruction output for a simulated eventwith a dual-phase detector geometry. Based on the experience with the same chain appliedto single-phase simulation the results can be judged as comparable, however standard effi-ciency test were not applied at this stage. Some tuning of algorithms may be expected toaccommodate differences in the two-point resolution or signal amplitudes over the long driftdistances in dual-phase detector geometry.
4 Non-beam physics
This section describes the progress in implementing simulation tools in LArSoft and adaptingthe standard reconstruction for the DUNE physics not related to the neutrino beam.
4.1 Supernova and Low-Energy Neutrinos
4.1.1 Nature of the Signal and Reconstruction Challenge
The DUNE Supernova Burst and Low-Energy (SNB/LE) Neutrino Physics Working Groupfocuses on physics that can be done with neutrinos of less than about 100 MeV. This isphysics enabled by the cosmogenically-quiet underground location of the detector. Theprimary physics topic is detection of the burst of neutrinos from a core-collapse supernova,with potentially very rich physics and astrophysics yield. The expected signal is a burst ofneutrinos of all flavors in the few- to few-tens-of-MeV range within tens of seconds, of whichthe component detectable by the DUNE LArTPC is primarily νe.
The SNB/LE group also considers solar neutrinos (energies up to ∼ 15 MeV) and thediffuse supernova neutrino flux (DSNB) which should have roughly similar properties tothe SNB flux, but much lower rate. For the latter two physics topics, the main issue is
21
LArSoft
Run:4 1/0Event:410
UTC4Thu4Nov46,4198012:13:25.162229104
t [ticks]9100 9200 9300 9400 9500 960010− 0
1020304050607080
350 400 450 500 550 600 650
9100
9150
9200
9250
9300
9350
9400
9450
9500
0 50 100 150 200 250 300 350
9000
9100
9200
9300
9400
9500
9600
9700
Figure 8: Example of the event reconstruction of a simulated proton with initial momentumof 2 GeV/c. The top two panels show event details in the channel number (x -axis) andtime tick (y-axis) for two collection views. The bottom panel shows an example of the hitreconstruction on the channels.
22
background. If background is well enough understood for burst neutrinos, then we shouldhave a handle on DNSB and solar neutrinos as well.
Supernova neutrino events, due to their low energies, will manifest themselves as small,perhaps few tens of cm, stub-like tracks from electrons (or positrons from the rarer ν̄e inter-actions). Events from νe charged-current interactions, νe+ 40Ar→ e−+ 40K∗, are likely to beaccompanied by de-excitation products– gamma rays and/or ejected nucleons. Gamma-raysare in principle observable via energy deposition from Compton scattering, which will showup as small charge blips in the TPC. Ejected nucleons may result in loss of observed energyfor the event. Elastic scattering on electrons will result in single scattered electrons, andsingle gamma rays may result from NC excitations of the argon nucleus. Each event categoryhas, in principle, a distinctive signature.
As yet, far detector requirements (on argon purity, wire pitch, triggering, time resolu-tion, photon collection, etc.) for supernova-neutrino detection are only roughly determined.Physics studies needed to determine these requirements require realistic simulations of signaland background and working reconstruction software.
The canonical reconstruction task is to identify the interaction channel, the neutrinoflavor for CC events, and to determine the 4-momentum of the incoming neutrino; thisoverall task is the same for low-energy events as for high-energy ones. The challenge is toreconstruct the properties of the lepton (if present), and to the extent possible, to tag theinteraction channel by the pattern of final-state particles.
So far, reconstruction algorithms for low-energy neutrinos in DUNE are poorly developedwith respect to those for high-energy events. While some physics study and detector require-ments work in the SNB/LE group uses a fast event-rate calculation tool called SNOwGLoBES,most activity is towards development of realistic and comprehensive simulation and recon-struction tools, from neutrino interaction event generators through full event reconstructionin both single and dual-phase detectors with LArSoft.
Understanding of backgrounds is also critical for understanding of how well we can recon-struct low energy events, and for setting detector requirements. Small single-hit blips from39Ar or other impurities may fake de-excitation gammas and also affect triggering. Back-grounds may be especially important for photon detectors (which in turn may be importantfor energy resolution by correcting for attenuation of charge during drift)— see Section 4.1.7.Work on backgrounds is done in collaboration with the DUNE Radiopurity group.
4.1.2 SNOwGLoBES
Most supernova neutrino studies done for DUNE so far, including of the plots included inthe CDR [30], have employed SNOwGLoBES[31], a fast event-rate computation tool. Thisuses GLoBES front-end software [1, 2] to convolve fluxes with cross-sections and detectorparameters. The output is in the form of interaction rates for each channel as a functionof neutrino energy, and “smeared” rates as a function of detected energy for each channel(i.e. the spectrum that actually be observed in a detector). The smearing (transfer) matri-ces incorporate both interaction product spectra for a given neutrino energy, and detectorresponse. Time dependence in SNOwGLoBES can be straightforwardly handled by providing
23
multiple files with fluxes divided into different time bins. 4
For default SNOwGLoBES, a detection threshold of 5 MeV in LAr is assumed as well asenergy resolution from Ref. [32]. Furthermore, the current default detector response smearingfor LAr in SNOwGLoBES assumes that all energy from de-excitation products is captured,which is likely a poor assumption. User-created transfer matrices have also been used forsome studies [33, 34].
While SNOwGLoBES is, and will continue to be, a fast, useful tool, it has limitationswith respect to a full simulation. One loses correlated event-by-event angular and energyinformation, for example. Ideally one does studies with high-statistics simulated data withrealistic event generators and detector response. When full simulation studies are constrainedby computational resources, one can update transfer matrices using improved simulation andstill use SNOwGLoBES to draw useful conclusions.
4.1.3 Event Generators
We have developed two event generators for supernova neutrino events.
• SNNueArCCGen is a simple νeCC event generator written by AJ Roeth and Gleb Sinev [35].This simulates νe charged-current interactions accompanied by de-excitation gammas.It calculates the cross section using a model based on measurements of 40Ti β+ de-cay to 40Sc, which is a mirror of the νe+
40Ar CC interaction’s 40K final state. Thismodel uses the beta-strengths to calculate the cross section and subsequently uses a73-level gamma de-excitation Monte Carlo to simulate the gammas. This generator isnow incorporated into LArSoft and samples from a cross-section weighted GKVM [31]spectrum by default. It does not take into account Ar excitations above 7 MeV, norfinal-state de-excitations via nucleons. However for some purposes, such as DAQ sim-ulations, this event generator should be satisfactory. This generator has been used toproduce samples for the MCC.
• MARLEY (Model of Argon Reaction Low Energy Yields), developed by Steven Gar-diner and the UC Davis group, supersedes SNnueArCCgen in physics modeling accuracy.MARLEY simulates tens-of-MeV neutrino-nucleus interactions in liquid argon. Cur-rently, MARLEY can only simulate charged-current νe scattering on 40Ar, but otherreaction channels will be added in the future.
MARLEY weights the incident neutrino spectrum, selects an initial excited state ofthe residual 40K∗ nucleus, and samples an outgoing electron direction using the allowedapproximation for the νe CC differential cross section.5 MARLEY computes this cross
4Note that SNOwGLoBES is not a Monte-Carlo code— it calculates mean event rates using a transfermatrix to convert neutrino spectra to observed spectra.
5That is, the zero momentum transfer and zero nucleon velocity limit of the tree-level νe CC differentialcross section, which may be written as
dσ
d cos θ=G2F |Vud|2
2π|pe|Ee F (Zf , βe)
[(1 + βe cos θ)B(F ) +
(3− βe cos θ
3
)B(GT )
].
In this expression, θ is the angle between the incident neutrino and the outgoing electron, GF is the Fermiconstant, Vud is the quark mixing matrix element, F (Zf , βe) is the Fermi function, and |pe|, Ee, and βe are
24
section using a table of Fermi and Gamow-Teller nuclear matrix elements. Their valuesare taken from experimental measurements at low excitation energies and a quasipar-ticle random phase approximation (QRPA) calculation at high excitation energies. Asthe code develops, a more sophisticated treatment of this cross section will likely beincluded.
After simulating the initial two-body 40Ar(νe, e−)40K∗ reaction for an event, MAR-
LEY also handles the subsequent nuclear de-excitation. For bound nuclear states, thede-excitation γ-rays are sampled using tables of experimental branching ratios. Thesetables are supplemented with theoretical estimates when experimental data are un-available. For particle-unbound nuclear states, MARLEY simulates the competitionbetween γ-ray and nuclear fragment6 emission using the Hauser-Feshbach statisticalmodel.
Although many refinements remain to be made, MARLEY’s treatment of high-lyingGamow-Teller strength and nuclear de-excitations represents a significant improvementover existing tools for simulating supernova νe CC events. An early version of MAR-LEY has been incorporated into the LArSoft code base and will soon be distributed inofficial LArSoft releases.
4.1.4 Report from Hack Days
On July 25-27, 2016, the SNB/LE working group held a “Hack Days” event at Fermilab,with the purpose of making progress on reconstruction tasks, and also to get group memberstrained on LArSoft and other software tools. This was very successful. Members signed upfor several tasks; some completed some shorter tasks and others were launched on longer-termprojects [36].
Table 4.1.4 lists some tasks and status. At the time of this report, most of the projectsare not complete. However we expect many to be completed, or significant progress made,on the timescale of a few months.
We plan to hold another Hack Days before the January 2017 collaboration meeting.
4.1.5 Reconstruction Progress
One very fruitful outcome of the Hack Days was development of a set of tweaks to thereconstruction in order to successfully reconstruct electrons with energies less than 20 MeV.The changes were suggested by Tingjun Yang and have been incorporated in the latestlarreco version. Figure 9 shows efficiency for reconstructing 3D space points before andafter these changes.
This first-order reconstruction of low-energy events is enabling many studies to go for-ward. However, we have yet to test effectiveness for reconstructing more complicated eventtopologies; identifying “blips” from secondaries will likely require additional algorithms.
the outgoing electron’s three momentum, total energy, and velocity, respectively. B(F ) and B(GT ) are theFermi and Gamow-Teller matrix elements.
6 Nucleons and light nuclei up to 4He are considered.
25
Figure 9: Efficiency for finding space points, defined as fraction of events with at least one 3Dspace point reconstructed, as a function of electron energy, for single electrons in the 10 ktworkspace geometry (fixed vertex and direction). The black line represents “out-of-the-box”LArSoft and the red line shows efficiency after improvements.
Figure 10: Example of reconstructed 10 MeV electron.
26
Person Task Status
C. Backhouse Study of photon detectors to improve Completeenergy reconstruction
E. Conley Study of bremsstrahlung vs Partially completede-excitation gammas
S. Gardiner MARLEY development LArSoft integrationcomplete,
studies underwayA. Habig Port of NOvA genie-based Partially completeJ. Vasel SN event generator
D. Navas Event reconstruction with MARLEY, Partially completesingle vs dual phase studies
C. Lastoria Angular resolution for ES channel Partially completeAJ Roeth Develop algorithm to select 10-MeV Reconstruction
NC gammas accomplished,selection complete
K. Scholberg New smearing matrices for SNOwGLoBES Partially completeusing realistic event generators
J. Stock Tl-208 impurity map Partially completeE. Stewart Simulated event production Partially complete
A. Weinstein for DAQ developmentD. Whittington New efficiency maps for determining Preliminary
t0 from photons for low-energy events maps created
4.1.6 DAQ and Triggering Progress
A key challenge for supernova detection will be to devise a way to trigger efficiently onthe burst and retain all relevant information. This is not trivial given that there is no beamtrigger, and the burst events are low enough in energy that required thresholds are low enoughthat data rates without clever zero suppression will be required to avoid overwhelming datarates. Strategies involve looking for bursts of stub-like events in the TPC over a time-scaleof 10 seconds, but in principle can incorporate also photon detector information.
The SNB/LE group is working with a subgroup of the DAQ working group (GeorgiaKaragiorgi, Amanda Weinstein, Michael Baird) to develop simulated data with characteris-tics suitable for trigger and DAQ studies. The short-term strategy is to use a fast and simplesimulator independent of LArSoft that generates trigger primitives based on the expectedenergy and time profile of a supernova burst. A first such time-profile simulator is underdevelopment by Michael Baird, and improvements will be informed by more detailed sim-ulation studies as they develop. Pile-up of burst signal events is not yet implemented butexpected to be small for all but the closest supernovae.
27
4.1.7 Photon Studies for Low-Energy Events
Photon information will also be important for supernova burst events for improving energyresolution by allowing for correction of drift time via the vertex information available fromfast photon timing. Furthermore they are in principle useful for triggering.
Preliminary simulation studies of photon detector efficiencies are shown in Fig. 11; ingeneral improved efficiencies would be desirable at low energy.
Figure 11: Left: photon simulation study result by Alex Himmel, evaluating efficiency ofphoton detection for different electron energies (red: 50 MeV, blue: 20 MeV, green: 10 MeV,black: 5 MeV) as a function of distance from light collection bar, for the DUNE referencedesign. Right: similar study for the alternate photon detector design.
So far, photon studies have included studies of energy resolution improvements. Fig. 12shows an example study by Gleb Sinev, demonstrating the resolution improvement (from∼ 22% to ∼ 13%) using MC truth to correct for electron drift. A Hack Days study by ChrisBackhouse demonstrated that a “realistic” correction using the photon simulation in placeof MC truth was nearly as good as the MC truth correction.
In addition, in collaboration with the radiopurity group, studies are underway of theeffect of 39Ar, and other backgrounds such as radon, on trigger rates (e.g., [37]): see Fig. 13.
A Photon Detector Task Force (see Section 6.4) will be examining these issues morecarefully in order to determine requirements for photon detectors, and will be working inclose collaboration with the SNB/LE group.
4.2 Nucleon Decay Physics
The purpose of the DUNE Nucleon Decay (NDK) Physics WG is to evaluate and demonstratethe experimental sensitivity of DUNE to various nucleon decay modes and other baryonnumber non-conserving processes, such as neutron-antineutron oscillations. Sensitivity to
28
Figure 12: Studies by Gleb Sinev of energy resolution for 15-MeV electrons for differentdrift times, without (top) and with (bottom) drift correction using MC truth position.
29
Figure 13: Study of 39Ar background for photon triggering by Gleb Sinev, showing efficien-cies and background rates for 8-MeV electrons, with simulated 39Ar. The dashed lines showthe effect of reduced background (with only modest decrease in efficiency) by requiring a2-flash trigger.
a specific NDK mode is expressed in terms of the partial lifetime τ/B, where B is thebranching fraction of the NDK mode being considered. Experimentally, the τ/B sensitivitycan be written as:
τ/B = n · ε ·Mt/NS (10)
where n is the number of protons or neutrons (depending on the NDK mode) per unitmass, ε is the signal selection efficiency, Mt is the detector exposure, NS is the largestnumber of signal events compatible with a background-only data sample. The lower thebackground rate, the lower NS, and hence the higher the τ/B sensitivity. DUNE sensitivitiesreported in the CDR [30] for the p → ν̄K+ mode, as well as in earlier studies, assumedpreliminary estimates for the signal efficiency and background rates [38]. An important goalof the NDK Physics WG, and of this Far Detector Task Force, is to refine these estimatesvia fully simulated and reconstructed event samples. Another important goal is to extendDUNE NDK studies to promising modes other than p → ν̄K+. This section focuses onrecent progress toward full simulation, reconstruction, and analysis for NDK signal events.The focus will be on p → ν̄K+, but other modes will be discussed as well. Cosmogenicbackgrounds to NDK are discussed in Sec. 4.4.
The simulation of NDK events includes the NDK event generation within genie [39, 40],the tracking of particles through the DUNE far detector geometry within geant4, and thehit-level simulation through LArSoft’s detsim module. The last two steps are common toother physics measurements and searches, and are discussed elsewhere. We focus here onNDK event generation.
genie is used for generation of nucleon decay events, since the treatment of nuclear effectson the initial and final nuclear state are common with the modeling of neutrino interactions,genie’s primary physics goal. During the last year, we have made major progress both
30
within genie, and concerning genie-LArSoft integration for NDK simulation. A summaryof the genie and LArSoft-related tasks accomplished by the NDK working group is listedhere:
1. Upgraded genie to include the full 68 exclusive nucleon decay modes listedby the Particle Data GroupThe current genie production release (v2.10) can only simulate 11 NDK modes,which are all (B − L)-conserving antilepton plus meson modes, see [39, 40]. However,the Particle Data Group reports experimental searches for as many as 68 exclusivemodes [41]. These modes include 2, 3, or 5 particles in the final state. They includeantilepton plus meson(s), lepton plus meson(s), antilepton plus photon(s), and three(or more) leptons decay modes. A genie Incubator Project to upgrade genie toinclude all such NDK channels has been created [42]. The code exists in a geniedevelopment branch and is under review by the genie team. The goal is to includethis upgrade in the genie v2.12 production release.
2. Improved the genie-LArSoft interface to allow seamless event generationwithout the awkward step that was previously required to bridge the twoformerly-independent pieces of software via an ASCII fileUntil recently, NDK event generation consisted of a LArSoft module reading genieevent records in ASCII format, assigning decay vertices with the argon of the detectorgeometry being used. This approach had the disadvantage that it required runninga stand-alone genie application. We are now able to simulate genie NDK eventsdirectly within the LArSoft module, through calls to the appropriate genie libraries,hence avoiding the need of a separate genie executable [43]. This greatly simplifies theproduction of NDK event samples as part of DUNE official Monte-Carlo productions.
3. Identified and fixed a bug in genie related to kaon inelastic interactionsIn genie (v2.10), kaon inelastic interactions have no particles in the final state forNDK-generated kaons, as in p→ ν̄K+. This has now been fixed, resulting in only 0.2%of K+ being lost within the Ar nucleus in p → ν̄K+ events, according to genie [44].This is to be compared with the 3.2% number in Ref. [38], based on the fluka [45]modeling of K+ + n→ K0 + p charge-exchange reactions within the Ar nucleus.
4. Identified and fixed a bug in LArSoft in which NDK event vertices wereincorrectly distributed through the argon volumeIn LArSoft, NDK vertices were incorrectly assumed to follow a gaussian distributionwithin the Ar volume. This has now been changed to follow a uniform distribution [46].Examples of fully simulated NDK events within the DUNE 1×2×6 far detector geom-etry are shown in Fig. 14. The left images in Fig. 14 show a p → µ+K0 event, apromising mode whose study in DUNE has been enabled by the recent simulationupgrades, while the right images show a p→ ν̄K+ “golden” mode event.
5. Upgraded genie to simulate neutron-antineutron oscillationA genie Incubator Project for neutron-antineutron oscillation simulation has beencreated [47]. This module is under review by the genie collaboration, and will also beincluded in a future genie production release.
31
z (cm)740 760 780 800 820 840 860 880 900
x (c
m)
20
40
60
80
100
120
z (cm)720 740 760 780 800 820 840 860 880 900
y (c
m)
20
40
60
80
100
120
140
160
180
p → μ+K0
μ+
e+
K0→π+π−
π−π+
z (cm)770 780 790 800 810 820
x (c
m)
8
10
12
14
16
18
20
22
24
26
z (cm)760 770 780 790 800 810 820
y (c
m)
265
270
275
280
285
290
295
300
p → νΚ+
Κ+
μ+
e+
Figure 14: Left: Event display of a fully reconstructed p → µ+K0 event simulated inDUNE, where µ+ → e+νeν̄µ and K0
S → π+π−. From [49]. Right: Event display of a fullyreconstructed p → ν̄K+ event simulated in DUNE, where K+ → µ+νµ and µ+ → e+νeν̄µ.From [50].
32
Progress has also been made in validating and improving reconstruction for NDK events.In addition to high tracking efficiency, excellent particle identification though the measure-ment of the dE/dx profile of stopping particles is crucial for NDK physics. DUNE aims notonly to efficiently reconstruct, but also to identify, all charged particles produced in NDKevents. The NDK working group has made progress in the following topics related to eventreconstruction and particle identification:
1. Evaluating tracking efficiencies for particles produced in p → ν̄K+ events,using the Projection Matching Algorithm (PMA, [27])For K+’s, the default PMA algorithm used for long-baseline oscillation physics studiesresults in a 64% overall tracking efficiency for K+’s [48]. The efficiency is highlymomentum-dependent, with an efficiency exceeding 80% for > 0.3 GeV/c kaons, asshown in the left panel of Fig. 15. In p → ν̄K+ events, kaons are expected to havemomenta distributed in the 0.1–0.5 GeV/c range, because although the decay producesa monoenergetic kaon, its momentum can be modified by Fermi momentum as the kaonescapes the nucleus. The efficiency for µ+ and e+ produced in K+ → µ+νµ decays(63.6% branching fraction) is not shown here, but is somewhat higher. We expect thatimprovements can be made during the next months by customizing the reconstructionalgorithms for better performance to relatively low momenta (∼0.1 GeV/c). As anexample, the left panel of Fig. 15 shows the results for K+ tracking efficiency using analternative cluster algorithm (TrajCluster module) compared to the standard algorithmfor tracks (LineCluster module). Tracking performance is also being evaluated in termsof single-track momentum and direction resolution, and for invariant mass resolutionfor two-body decays of neutral mesons (π0, K0
S, ρ0) [48, 49].
2. Evaluating PID efficiencies and mis-ID rates for e/µ/π/K/p separation [49,50]Distributions from a particle identification algorithm (PIDA, [51]) applied to all recon-structed particles in simulated p→ ν̄K+ events is shown in the right panel of Fig. 15.The algorithm is applied here only on the last 15 cm of the tracks. Studies are inprogress to understand the current status and improve where possible. An importantdifference between beam and NDK events is that no preferred direction exists in thelatter case, and detector-only information can be used to determine the track direc-tion. We have verified with (p → ν̄K+, K+ → µ+νµ) events that the same dE/dxinformation along the track works very well also to determine the K+ track direction,and hence the NDK vertex [50].
3. Assessing photon detector performance in relation to NDK physics [52]The photon detector system (PDS) is sensitive to the primary scintillation light fromAr de-excitation and electron-ion recombination. The PDS is particularly importantto enable non-beam physics discussed here, especially to provide drift position deter-mination and hence fiducial volume definition. About 103 photo-electrons (PEs) areexpected on average for p → ν̄K+ events and the current PDS assumptions. Theamount of detected light varies greatly with the distance between the NDK vertex andthe APA plane where the PDS is installed. Figure 16 shows the efficiency to recon-struct a flash of light as a function of such distance, and for different PE thresholds for
33
Table 1: List of possible NDK modes where DUNE may have an early discovery potential.See text for details. The current limits for these modes are also shown in the table. Adaptedfrom [53].
Mode PDG ID Current Limit (1033 yr)n→ e+K− 13 > 0.017p→ e+K0 13 > 1.0n→ µ+K− 16 > 0.026p→ µ+K0 16 > 1.6p→ e+K∗0 21 > 0.084n→ e−K+ 34 > 0.032n→ µ−K+ 35 > 0.057p→ e−π+K+ 40 > 0.075p→ µ−π+K+ 41 > 0.245
the PDS (in the 2–20 PE range). The efficiency is satisfactory: it is greater than 97%for NDK vertex-APA distances lower than 340 cm. Fake associations between chargeand light signals might be produced by light flashes from 39Ar radioactive decays. Thenumber of reconstructed 39Ar flashes can be significant for few-PEs thresholds, as alsoshown in Fig. 16. In the future, the spatial reconstruction of the flash across sev-eral PDS elements will be studied to discriminate flashes of light between 39Ar, NDK,atmospheric neutrino and cosmogenic events.
9
Proton Decay, p → K+ ⌫̄
Kaon Momentum (GeV)0 0.1 0.2 0.3 0.4 0.5 0.6
Tra
ckin
g Ef
ficie
ncy
0
0.2
0.4
0.6
0.8
1
Michele e Momentum (GeV)0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Tra
ckin
g Ef
ficie
ncy
0
0.2
0.4
0.6
0.8
1
Muon Momentum (GeV)0 0.1 0.2 0.3 0.4 0.5 0.6
Tra
ckin
g Ef
ficie
ncy
0
0.2
0.4
0.6
0.8
1
There seems to be room for improvement with some tweaks
physics.producers.pmtrack.AutoFlip_dQdx: true physics.producers.pmtrack.MinSeedSize2ndPass: 2
Uses trajCluster as input instead of LineCluster
5
Proton Decay, p → K+ ⌫̄
For now we are going to focus only on K → µ+ + vµ events!PIDA of the last 15 cm of track
PIDA0 5 10 15 20 250
200
400
600
800
1000
1200
1400
1600
PIDAi = (dE/dx)caloR0.42
Kaon!Muon+!Michel e+!Proton!Others
Figure 15: Left panel: K+ tracking efficiency versus K+ momentum in p → ν̄K+ events.Blue symbols: using standard 2D clustering algorithm (LineCluster) as tracking input. Blacksymbols: using alternative clustering (TrajCluster). Right panel: PIDA distribution forreconstructed particles in p→ ν̄K+ events where K+ → µ+νµ. PIDA is computed using thelast 15 cm of the tracks. From [50].
As far as NDK analysis, the focus has been on:
1. Identification of promising NDK modesA comprehensive list of NDK modes where DUNE may have an early discovery po-tential has been produced [53], see Tab. 1. In this context, early discovery potential is
34
Threshold 39Ar flashes(PEs) (#/APA/drift)
2 305 310 0.0815 0.00420 0.0003
Figure 16: Flash finding efficiency for p → ν̄K+ events, as a function of the flash distancefrom the closest APA and for different PE thresholds. The table at the right indicates theexpected rate of 39Ar flashes above threshold. From [52].
defined as DUNE having a sensitivity after a 40 kton·yr exposure that is potentiallygreater than the current limit and the Super-Kamiokande sensitivity extrapolated tothe year 2025. The assumed DUNE sensitivities account for nuclear effects responsiblefor irreducible efficiency losses. However, they should be regarded as rough indicationsonly, as they assume zero background and approximate reconstruction efficiencies.Nonetheless, they provide a useful subset of modes that are candidates for more real-istic Monte-Carlo simulations. As the table shows, DUNE is expected to do well inNDK modes with charged or neutral kaons in the final state, or in multi-prong decays.Modes with neutrinos in the final state may also be promising. They are, however,not listed in Tab. 1, as reliable estimates for their relatively large backgrounds areparticularly important in this case.
2. Evaluating NDK signal efficienciesA realistic estimate of NDK signal efficiencies for fully reconstructed p → ν̄K+, p →l+ρ0 and p→ µ+K0 events has just started [49, 50], and will continue to make progressin coming months.
3. Evaluating NDK background ratesA realistic estimate of NDK backgrounds from both cosmogenically-induced events andfrom atmospheric neutrino events is a critical component of these analyses. Cosmogenicbackground studies are discussed in more detail Section 4.4. Studies of atmosphericneutrino backgrounds are just beginning.
4. Evaluating NDK sensitivities
35
Python-based scripts have been developed to evaluate DUNE sensitivities as a functionof exposure, signal efficiency, and background rate assumptions. These are availableto the collaboration [54].
The NDK working group has made good progress over the last year in each of the areasthat is critical for the success of NDK searches in DUNE. Although there is still much workto be done, the forward motion over the past year is largely thanks to the efforts of ourdedicated working group members.
4.3 Atmospheric Neutrino Physics
Atmospheric neutrino studies for LBNE and LAGUNA-LBNO relied on smeared truth-levelquantities rather than full simulations and reconstruction. These previous studies focused onthe neutrino mass hierarchy, determining the octant of θ23, and observation of CP violation.In DUNE the focus of effort has been to continue the study of interesting physics signatures,and to move away from truth-level studies to full simulation and reconstruction.
Atmospheric Neutrino Simulation and Reconstruction: Within LArSoft, userscan now generate and process atmospheric neutrino events through all of the necessarystages: event generation with genie, geant4 simulation of energy deposition, detectorsimulation, reconstruction, and analysis. The current LArSoft configurations allow one tosimulate events using the Bartol atmospheric neutrino fluxes calculated at solar max or solarmin [55]. Improvements have been made to numerous elements of this software stack, andareas where further work is necessary have been identified.
The generation stage brings together the detector mass model, an atmospheric neutrinoflux calculation, and the neutrino interaction cross sections, using a piece of genie codeknown as the Event Generation Driver. Several groups world-wide carry out calculations ofatmospheric neutrino fluxes [55, 56, 57], and the genie flux drivers have been extended sothat they are able to use calculations from all of these groups. Most of these groups have nowperformed calculations for the geomagnetic location of the Homestake site, and we hope inthe future to also have calculations which include the local topography [58]. The addition ofthe ATMNC flux [57] necessitated changes to the structure of the genie atmospheric eventdrivers, as this computation included additional information, such as the azimuthal-angledependence of the flux, and direction-dependent production height tables, which were notincluded in previous flux calculations. Some additional work will be required to extend thegenie output record in order to pass through needed flux-related information, such as theneutrino production height, for the purposes of downstream analyses. The existing driversare rather slow - arising from the fact that atmospheric neutrinos have a broad and rapidlyfalling energy spectrum, and by default genie strobes neutrinos through the full detector inorder to compute the probability of interaction (relative to the maximum probability whichit also determines). This has not proved to be a bottleneck for analyses studies to date, butis an area where significant speedup is possible.
There has also been improvement on the cross section side of genie with the incorporationof the Athar et al. model for neutrino ∆S = 1 scattering into genie release 2.10.0 [59, 60].This model, which provides an improved description of ∆S = 1 CC interactions up to 3GeV, is not turned on by default, but is expected to be part of a comprehensively retuned
36
Figure 17: Muon tracking efficiency for atmospheric muon neutrino CC interactions as afunction of neutrino energy (left) and muon momentum (right). From [48].
version of genie available in the future.Atmospheric neutrino interactions have been carried through the entire simulation and
analysis framework, and preliminary studies on how well the reconstruction works on theseevents have been performed [48]. Figure 17 shows the muon tracking efficiency for atmo-spheric muon neutrino CC interactions as a function of neutrino energy and muon mo-mentum. Overall the muon tracking performance is found to be similar to that for beamneutrinos, with tracking failures most often occurring when muons are traveling parallel tothe electron drift. As the atmospheric neutrino beam is arriving from all directions thisparticularly challenging orientation can be achieved for any muon (or neutrino) energy. Un-derstanding the energy and angular dependence of these efficiency curves will be importantfor atmospheric neutrino studies which rely on resolving features of neutrino oscillograms.
The previous studies demonstrated that DUNEs excellent neutrino energy and pointingresolution can compensate for the relatively small fiducial volume, resulting in competitivephysics sensitivities. The mass hierarchy sensitivity, for instance, relies primarily on mea-surement of neutrinos with energies from 1 to 10 GeV, an energy regime largely insensitiveto some of the detector optimization decisions currently being debated. Over the past yearwe have therefore tried to encourage study of additional DUNE physics topics that mighttake advantage of these same strengths for lower energy interactions. Two examples are thestudy of matter effects using low energy atmospheric neutrinos [61], and the measurement ofthe neutrino azimuthal energy dependence [62]. The measurement of the azimuthal depen-dence of atmospheric neutrinos has been carried out by the Super-Kamiokande experiment[63], and a comparable measurement by DUNE could potentially be done with a significantlysmaller data set and serve as a valuable early demonstration that the detector is capableof achieving high resolution measurements. This validation would be particularly impor-tant as more data is acquired and we attempt to discern other features from the measuredflavor-energy-angle dependent oscillograms. For a particular bin of zenith angle, flavor, andenergy, neutrino oscillation probabilities are independent of neutrino azimuthal angle, andflux calculations predict significant non-isotropies resulting from the geomagnetic shieldingof low energy primaries. Figure 18 shows the ratio of Honda and Bartol flux predictions formuon neutrinos, as a function of zenith angle, azimuthal angle, and energy.
37
(deg)φ
0 50 100 150 200 250 300 350
)/0.1
θB
art
ol/H
onda +
cos(
10
5
0
5
10
flux ratiosµνBartol/Honda flux ratiosµνBartol/Honda
Figure 18: Ratio of the Honda to Bartol flux calculation for muon neutrinos. Each bandcorresponds to the flux in a zenith angle bin of cos(θ) = 0.1, with downgoing neutrinos inthe top band. Within each band are ten colored lines corresponding to increasing neutrinoenergies in 1 GeV bins from 0 to 10 GeV. From [62].
38
In this calculation, the flux driver used for the Bartol flux did not include azimuthalinformation, so the flux was taken to be isotropic in azimuthal angle. The plot demonstratesseveral features the East/West asymmetry for incoming neutrinos, and the relatively largefluxes of low energy neutrinos arriving from the Earths polar regions (as seen from a detectorat the Homestake site).
This results from the fact that the magnetic shielding of the Earth is smallest at the poles.Figure 19 shows the true azimuthal angle distribution for atmospheric neutrino interactionswith cos(θ) < −0.7, a region covering the South pole, from a sample of 500k simulatedatmospheric neutrino interactions. The black points correspond to events with Eν < 0.5GeV, red to Eν > 3 GeV, clearly showing the large variation for low energy neutrinos. Figure20 shows the expected DUNE azimuthal angle resolution for Eν < 1 GeV interactions, whichis clearly sufficient for resolving these expected features in the atmospheric flux.
Azimuthal angle (rad)
-2 0 2
Eve
nts
0
100
200
300
400
500
Figure 19: True azimuthal angle distribution for atmospheric neutrino interactions withcos(θ) < −0.7, from a sample of 500k simulated atmospheric neutrino interactions [64].Eν < 0.5 GeV (black) and Eν > 3 GeV (red).
4.4 Cosmic Rays and Cosmogenics
4.4.1 Muon generator: MUSUN
The muon generator for the DUNE far detector (FD) is based on the well-known muon sim-ulation code MUSUN (MUon Simulations UNderground) [65, 66]. The code uses the results
39
Figure 20: Azimuthal angle resolution for νµ CC interactions with Eν < 1 GeV [62].
of muon transport through rock carried out with the MUSIC code [67, 68, 65] convolvedwith the muon spectrum at sea level for different zenith angles. The MUSUN code, origi-nally written as a standalone package in FORTRAN, has been converted into a module inLArSoft (written in C++) and tested with respect to the original code.
The MUSUN code, as currently adapted for DUNE, samples muons around an under-ground laboratory located near the Ross shaft and expected to host one of the 10 kt mod-ules of the future DUNE experiment. The global coordinates of the location: latitude =44◦20′45.21” N, longitude = 103◦45′16.13” W. The rock composition has been taken fromRefs. [69, 70]. Many rock samples have been measured and the average rock composition hasbeen calculated as < Z >= 12.09 and < A >= 24.17 [69, 70]. The average density of rockwas assumed to be 2.70 g/cm3 [70] in the MUSIC simulation for SURF. Other measurementssuggest that the density may be larger (2.8-2.9 g/cm3 [71, 72]). The density can be changedin the MUSUN muon generator if required. The slant depth distribution for the 4850 ft levelat SURF was calculated from the CGIAR digital elevation model [73].
The measurement of the muon flux at SURF (in the Davis’ campus near the Yates shaft)was performed by the active veto system of the Davis’ experiment [74] giving the value of(5.38±0.07)×10−9 cm−2s−1sr−1 for the vertical flux (the measured fraction of multiple muonshas been added to the single muon flux as given in Ref. [74], so the systematic uncertaintyfor this value may be of the order of a few %). The result agrees very well with the verticalflux calculated by MUSIC/MUSUN (in this case the coordinates for the Davis’ campus wereused to calculate the slant depth distribution for MUSUN): 5.18 × 10−9 cm−2s−1sr−1. Arecent measurement of muons in the Davis’ campus has been carried out with the activeveto system of the Majorana demonstrator [75] giving the value of (5.04 ± 0.16) × 10−9
cm−2s−1 for the total (not vertical) muon flux, in reasonable agreement with a calculationwith MUSIC/MUSUN for a close location (also in Davis campus but LUX/LZ location ratherthan Majorana demonstrator hall): 6.16× 10−9 cm−2s−1. Given a small difference between
40
the recent measurement and simulations (that can be due to different locations) we estimatethe systematic uncertainty in calculating the muon flux in our model as 20%. We emphasizethe need for an accurate measurement of the muon flux in the DUNE cavern(s) to normalizesimulations.
Figure 21 (left) shows the profile of the surface above the proposed location of the fardetector (center of the map). The lines drawn from the center divide each quadrant into 4angles of equal size, 22.5◦, to guide the eye. Muon azimuth angle distribution is plotted inFigure 21 (right). The vertical lines show approximately the division of quadrants on theleft figure where the azimuth angle is calculated from East (pointing to the right on the leftfigure). Moving from East to North and further on counterclockwise on the map on the left,one can see how the dips and peaks on the surface profile correspond to peaks and dips inthe number of muons on the graph on the right.
Figure 21: Left – surface profile above the proposed location for the far detector (centerof the map). The lines drawn from the center divide each quadrant into 4 angles equal insize: 22.5◦ to guide the eye. Right – muon azimuth angle distribution. The vertical linesshow the division of quadrants on the left figure where the azimuth angle is calculated fromEast (pointing to the right on the left figure). Moving from East to North and further oncounterclockwise on the map on the left, you can see how the dips and peaks on the surfaceprofile correspond to peaks and dips in the number of muons on the graph on the right.
The parameters of the muon flux are given in Table 2 for simulations carried out withMUSIC/MUSUN. These particular simulations did not include the cavern geometry andassumed spherical geometry for an imaginary detector.
The normalization of the muon flux or rate is done for a specific point undergroundwhere the slant depth distribution was calculated. This implies that the cavern/laboratoryis relatively small and the muon flux is the same everywhere across the cavern (plus the rockwhere the box for muon sampling is extended to). For the cavern(s) where the far detectorwill be located the difference in muon intensities in different locations does not exceed 10%.
41
Cosine of zenith angle0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Num
ber
of e
vent
s
1
10
210
310
410
510
Muon energy (GeV)1 10 210 310 410 510
Num
ber
of e
vent
s pe
r G
eV
5−10
4−10
3−10
2−10
1−10
1
10
210
310
Figure 22: Left – zenith angle distribution of muons as simulated with the MUSUN muongenerator for DUNE, right – energy spectrum of muons at the DUNE site.
Depth, km w.e.4 6 8 10 12 14
Num
ber
of e
vent
s
1
10
210
310
410
510
610
Zenith, degrees0 10 20 30 40 50 60 70 80 90
Azimuth, degrees 050
100150
200250
300350
Num
ber
of e
vent
s
1
10
210
310
Figure 23: Left – distribution of muons according to the distance to the surface that theyhave travelled through, right – distribution of the number of muons as a function of zenithand azimuth angles.
Zenith, degrees0 10 20 30 40 50 60 70 80 90
Azi
mut
h, d
egre
es
0
50
100
150
200
250
300
350
Num
ber
of e
vent
s
1
10
210
310
X (cm)2000− 1000− 0 1000 2000
Y (cm)
2000−0
20004000
6000
Z (
cm)
2000−
1000−
0
1000
2000
Figure 24: Left – Colour plot of the number of muons as a function of zenith and azimuthalangles, right – muon positions as sampled on the surface of the box around the cavern withthe cryostat.
42
Table 2: Muon flux parameters as calculated with MUSIC/MUSUN.
Total flux, cm−2s−1 Mean Eµ, GeV Mean slant depth, m w. e. Mean θ, deg.5.66× 10−9 283 4532 26
In the MUSUN generator for DUNE muons are sampled on a surface of a box that en-compasses the lab, about 7 m of rock above the cavern and 5 m of rock beside and belowthe cavern. The rock is included to make sure that muon-induced cascades initiated aboveand around the cavern, have enough of space to fully develop and produce, in simulations,potentially dangerous secondaries, whereas the muon itself may not cross the active volumeof argon. The size of the box is 74.43×29.54×30.18 m3 (length×width×height). The size ofthe cryostat has been assumed to be 61.62×14.94×13.58 m3 (length×width×height). Muonsare sampled according to their energy spectrum for a particular zenith and azimuthal angles,sampled in its turn from the angular distribution obtained using the MUSIC code. The sizeof the box (the probability of a muon to cross a particular surface of the box) is taken into ac-count when generating muons. All energy spectra and angular distributions are stored on diskand read during the initialization at the beginning of each simulation run. The generator isfully integrated into larsim (within the LArSoft built upon art framework) since v04 24 00.The module is located at larsim/larsim/EventGenerator/MuonPropagation/MUSUN module.cc.The configuration fcl file is in the same directory and is called MUSUN.fcl. MUSUN is runwith the fcl file prodMUSUN dune10kt.fcl.
The muon rate through the surface of this box is 0.1579 Hz. This rate is used laterto normalize the muon-induced background event rate in DUNE. (Note that the overalluncertainty in the muon flux is about 20%, see above.) The rate of muons passing throughthe active volume of liquid argon is 0.053 Hz so about 30% of muons generated on the surfaceof the box are passing through active argon (in fact having the track length of more than1 m in the active volume.
Figures 22, 23 and 24 show various distributions of muons obtained with the MUSUNgenerator as incorporated into larsim.
4.4.2 Muon background for proton decay
There are a large number of proton decay modes where the sensitivity of DUNE may becomparable to or even exceeding that of Super-K/Hyper-K. The mode which is usually con-sidered to be the ’golden’ mode for the discovery at DUNE is: p → K+ν̄. This mode isdominant in most supersymmetric GUTs, many of which also favor other modes involvingkaons in the final state. The decay modes with a charged kaon are unique for LAr experi-ments; since stopping kaons have a higher ionization density than pions or muons with thesame momentum, a LArTPC could detect them with extremely high efficiency. In addition,many final states of K+ decay would be fully reconstructable in a LArTPC. See Table 1 fora list of the promising nucleon decay modes.
The key signature for p→ K+ν̄ is the presence of an isolated charged kaon (which wouldalso be monochromatic for the case of free protons, with the momentum p ≈ 340 MeV).
43
The kaon emerges intact from the nucleus with 97% probability. The kaon momentum issmeared by the proton’s Fermi motion and shifted downward by re-scattering [76]. Thekaon emerging from this process is below Cherenkov threshold in water, therefore a waterCherenkov detector would need to detect it after it stops, via its decay products. In LArTPCdetectors, the K+ can be tracked and identified via detailed analysis of its energy loss profile,and its kinetic energy measured by range. Additionally, all decay modes can be cleanlyreconstructed and identified, including those with neutrinos, since the decaying proton isessentially at rest. With this level of detail, it is possible for a single event with an isolatedkaon of the right momentum originating from a point within the fiducial volume, to provideoverwhelming evidence for an observation of a proton decay.
The background for a proton decay search comes from atmospheric muons and neutrinos.If a muon passes through the detector, then the event can easily be reconstructed as abackground muon-induced event and rejected. The current design of DUNE far detectorsuggests that the rate of muons passing through active LAr of a single module is about 0.05Hz. Assuming a maximum drift time (∼few ms) for an event, rejecting all events with muonscrossing the active volume of the TPC will result in less than 0.1% of dead time. Thus, onlyevents with muons not crossing the active LAr volume may contribute to the background.Among these events the main danger comes from neutral kaons produced outside the TPCand undergoing inelastic scattering (primarily charge exchange) inside the TPC resulting ina single positive kaon detected. Figure 25 shows a schematic for such background event.Potential cosmogenic backgrounds with other particles mis-reconstructed as kaons have notyet been assessed, and are pending improvements in reconstruction algorithms.
Figure 25: A schematic showing a potential background events with a muon evading thedetector but producing a neutral kaon that produces a positive kaon in its turn via chargeexchange reaction. Positive kaon decay products are not shown. Taken from [73].
We have carried out initial simulations of muon-induced background in one DUNE modulewith a fiducial mass of about 10.4 kt. By now (August 2016) 4 × 108 muons have beensimulated and events in the detector analysed, corresponding to 80.32 years of live time ofrunning one DUNE module. Only ’truth’ parameters have been processed due to CPU timelimitations and the full reconstruction chain not being fully ready for reconstructing largecosmic-ray events. We start by identifying potential backgrounds as events with a chargedkaon in the TPC active volume (positive and negative kaons cannot easily be separated in
44
LAr) whereas a primary muon misses the detector. The following cuts have been applied toreject muon-induced events: (i) no muon is in the detector active volume (the track length ofthe initial muon is less than 20 cm), (ii) the K+ is fully contained within the fiducial volume(> 2 cm from any TPC wall, similarly no energy deposition with 2 cm from the walls),(iii) the energy deposition from the K+ and its descendants (excluding decay products)is < 250 MeV (this includes smearing of the energy deposition due to energy resolutionand nuclear effects in proton decay), (iv) the total energy deposition from the K+, itsdescendants and decay products is < 1 GeV, (v) energy deposition from other particles inthe muon-induced cascade (i.e. excluding the energy deposition from the positive kaon, itsdescendants and decay products) is < 30 MeV (otherwise this additional energy depositioncan be clearly identified as such and the event would be rejected as a background).
Figure 26 (left) shows the spectra of energy depositions from charged kaons after thediscussed cuts are applied. The three events shown by a dashed blue histogram are rejectedafter the additional cut on the energy deposition from other particles is applied. Figure 26(right) shows the scatter plot of the kaon energy deposition versus that of other particles(excluding kaon and its decay products) for background events. For a proton decay eventthe energy deposition of other particles (x-axis on this plot) should be 0 but we have allowedfor some hits to be missed or attributed to radiogenic background, hence imposed an upperlimit of 30 MeV on this energy deposition. The 3 events shown as red circles on the rightplot (also shown as dashed blue histogram on the left plot) are rejected because the energydeposition from other particles exceeds 30 MeV. A few events inside the region of interest arerejected because of the energy deposition within 2 cm of the walls (particles are entering theactive volume from outside). Note that rejecting some of these events require good timingresolution (light detection) to identify the correct start of the event. The study of the protondecay detection efficiency with a current design of the DUNE far detector is underway. Ourprevious studies with a simplified cylindrical geometry of the cryostat show that 10 cm cutaround the walls reduce the detection efficiency by less than 2% [77]. As a result of theanalysis no event survived all the above cuts, resulting in an upper limit (at 90% CL) on themuon-induced background rate of 0.0029 events/kt/year.
Figure 27 (left) shows the 2D-distribution of simulated events from proton decay modep → K+ν̄. The x-axis shows the energy deposition from kaon decay products whereas they-axis shows the energy deposition from the positive kaon. Almost all kaons deposit lessthan 250 MeV inside the active volume which justifies the choice of cuts for backgroundevents. Note that the energy deposition from other particles (excluding kaons and theproducts of their interactions and decays) should be zero for this mode of proton decaywhich serves as a good selection criterion to reject background events (currently required tobe less than 30 MeV but can be tuned further). Figure 27 (right) shows a similar distributionfor cosmogenic background before the discussed cuts are applied. The distribution is differentfrom that for proton decay events due to the fact that background kaons are mostly enteringthe active volume from outside.
Simulated cosmic-ray events are stored on disk and can be analyzed to determine potentialbackground for other nucleon decay modes. They will also be processed further through thewhole data collection, track reconstruction and particle identification chain.
45
MeV1 10 100 1000 10000 100000
/MeV
3−10
2−10
1−10
1
10
Total energy deposition (42309)
With Kaon track (2993)
and One Kaon track in event (1011)
and No Muon track >20 cm (566)
and No hits within 2 cm of walls (3)
Other energy deposition [MeV]1 10 210 310 410 510
ene
rgy
depo
sitio
n [M
eV]
+/-
K
1
10
210
310
410
Reg
ion
of in
tere
st
Passing all but fiducial and energy cuts (563)
Also passing fiducial cut (3)
Figure 26: Left – spectra of energy depositions from kaons after various cuts, right – scatterplot of a kaon energy deposition vs energy deposition of other particles (excluding kaon andits decay products) for background events.
4.4.3 Cosmic-ray event reconstruction
Reconstruction in LArTPCs is a complex process requiring numerous steps from hit recon-struction to hit disambiguation, clustering and ultimately combining clusters into tracks.As the process has multiple steps there are a suite of options at each stage, for this studythough the following modules have been used:
• Gaussian hit finder,
• Disambiguation algorithm tuned for the far detector,
• Line cluster, clustering algorithm,
• Projection matching tracking algorithm (PMTrack).
Many of the reconstruction algorithms were tuned and developed for significantly smallerdetectors than the far detector, such as the 35 ton prototype. This means that tuning of thealgorithms is required to achieve a high reconstruction efficiency. This tuning was achievedthrough a detailed study of 100 simulated muons using the MUSUN generator discussedabove. Properties of the reconstruction to be improved were identified and the tuning re-quired to do so is illustrated below.
Two views are shown for each event, one called ’Ortho3D’ which has two dimensionalrepresentations of the detector in the XZ and YZ planes. Within each plane the TPCs areshown as regions contained within gray dashed lines. The reconstructed tracks are shownas bold colored lines, and are labeled with their reconstructed track number which increasesfrom 0. The second view is that of the wire planes for a given TPC. The TPC number isshown on the left hand side of the image. Three wire planes are visible, which in descendingorder are the collection plane, the U induction plane and the V induction plane. The plotfor each plane shows the wire number on the plane (increasing from 0) against TPC tick (in
46
0
5
10
15
20
25
30
35
decay products [MeV]+/-Energy deposition from K0 50 100 150 200 250
Ene
rgy
Dep
ositi
on [M
eV]
+/-
K
0
20
40
60
80
100
120
140
160
180
200
Figure 27: Distribution of simulated events from proton decay mode p→ K+ν̄. The x-axisshows the energy deposition from kaon decay products whereas the y-axis shows the energydeposition from the positive kaon.
units of 500 ns). The charge deposited on a wire at a given time is shown using a color scaleshown on the right of each plane. At the bottom of the image is a plot showing the chargedeposited on a given wire for a range of times, the blue line shows the de-convoluted signaland the orange lines displays the reconstructed hits.
Many through-going muons were originally reconstructed as multiple tracks as opposedto a single track. These track separations regularly occurred as a muon crossed a TPCboundary, where there is an ∼5 cm area of un-instrumented LAr. This meant that the tracksections were not merged into a single track. Upon allowing for tracks to be merged withlarger separations many of the un-merged tracks are observed to be successfully merged. Anexample of a merged through-going muon is shown in Figure 28.
Another reason for split tracks is due to high energy delta-rays. These large energy depo-sitions create additional large clusters to be reconstructed around the main track causing thetrack to be split. Further study is required to mitigate the effect of high-energy delta-rays,an example of which is shown in Figure 29.
High-energy muons can produce large showers. These large energy depositions causemany short (≤10 cm) tracks to be reconstructed. The effect of this is that the reconstruc-tion appears to perform badly due to a large excess of reconstructed tracks at short tracklengths, however the through-going muon track is often fully reconstructed. An example ofa high-energy shower is shown in Figure 30.
Occasionally hits within TPCs can be successfully reconstructed into clusters which arenot merged to produce tracks. It was observed that this is due to one or more of the clusters
47
Figure 28: An example of a well reconstructed through going muon track. Left image showsthe Ortho3D view, right image shows the LArSoft wire plane display.
being incorrectly identified as a delta-ray causing the track to be reconstructed outside theTPC and being discarded. The effect of this was greatly reduced by allowing for a greatervariation in TPC containment, however occasional examples can still be found. An exampleof an un-merged track is shown in Figure 31.
Reconstruction of kaons of the energy which would be observed in proton decay is beingstudied by the reconstruction working group and is not considered in here.
4.4.4 Cosmogenics and Requirements for the Far Detector
Cosmic rays impose the following design requirements for the DUNE far detector modules:
1. Each 10 kton module will detect cosmic-ray muons at a rate of ∼0.05 Hz. The resultingsignals are expected to dominate the data collected by the DUNE far detectors, andwill therefore set requirements on the average data bandwidth and total data storage.
2. Cosmogenic backgrounds are of most concern to searches for nucleon decay. The workpresented here investigates methods to reject these backgrounds. The study impliesthat good timing resolution for the start of each event, O(10 µs) or better, is necessaryin order to properly define the fiducial volume for nucleon decay. Reducing potentialdead volumes in the TPC is useful, although it is not as critical. A potential backgroundfrom mis-reconstruction of cosmogenic backgrounds (without true kaons) as nucleondecay candidates requires additional study. Cosmogenic backgrounds for nucleon decaymodes beyond the ’golden’ mode (p→ K+ν) require further study.
3. Cosmic ray muons will likely serve a role in calibration of the DUNE far detectors,but it is not yet clear whether this imposes any addition detector design requirementsbeyond those already considered.
48
Figure 29: An example of a delta-ray causing a through-going muon to be split into twotracks. Left image shows the Ortho3D view, right image shows the wire plane display.
5 Long-baseline Physics Sensitivity Studies
5.1 Event Selection
A critical aspect of the long-baseline oscillation analysis is identification of the νµ and νeevent samples. In the Fast Monte Carlo, which is currently used to determine the signaland background selection efficiencies used in GLoBES-based sensitivity calculations and asinput to the DUNE oscillation fitters described below, the selection criteria are applied toreconstructed quantities, which are the result of applying a parameterized detector responseto generator-level particles, at the individual particle level. For the full LArSoft Monte Carlo,a multivariate analysis has been developed to perform the νe and νµ signal selection using(mostly) reconstructed quantities.
The Fast MC event selection is fairly simple. Any event with a candidate muon, defined asa track longer than 2 m, is classified as νµ CC. Any event without a candidate muon, but witha candidate electron/positron, which is defined as an electromagnetic shower (EM) that isnot consistent with a photon, is classified as a νe CC. An EM shower is classified as consistentwith a photon if it begins 2 cm or more from the event vertex, or if it can be matched withanother EM shower to reconstruct an invariant mass with 40 MeV of the π0 mass. In a fullanalysis, the dE/dx of the initial portion of shower can discriminate between photon-likeand electron-like showers; the efficiency for this selection is applied probabilistically in theFast MC as a function of shower energy and hadronic-shower multiplicity, based on resultsfrom early Monte Carlo studies. Finally, to reduce background from neutral current and ντin the νe CC-like sample, an additional discriminant is formed using reconstructed transversemomentum along with reconstructed neutrino and hadronic energy as inputs to a k-Nearest-Neighbor (kNN) machine-learning algorithm. Figure 5.1 shows the analysis sample detectiontimes selection efficiencies for the various signal and background modes extracted from theFast MC.
49
Figure 30: An example of a high-energy shower in the detector around a through goingmuon. Left image shows the Ortho3D view, right image shows the wire plane display.
To move beyond the probabilistic event selection used in the Fast MC, a multivariateanalysis has been developed to perform νe and νµ event selection in LArSoft. νµ CC eventsare identified using a TMVA algorithm with input from the following variables: numberof tracks, maximum track length, average track length, track dE/dx, signal fluctuation,transverse track profile, maximum fraction of charge in 5, 10, 50, and 100 wires, directioncosines of longest track, and fractional transverse energy. νe CC events are identified usinga TMVA algorithm with input from the following variables: number of tracks, number ofshowers, maximum track length, average track length, track dE/dx, shower dE/dx, signalfluctuation, transverse track profile, maximum fraction of charge in 5, 10, 50 and 100 wires,direction cosines of the longest track and the highest energy shower, distance from neutrinovertex to shower vertex, fractional transverse energy, number of hits per wire in the shower,and shower length. The variables considered by the multivariate analyses are all extractedfrom automated reconstruction routines applied to DUNE LArSoft Monte Carlo, with theexception that “disambiguation cheating,” which makes use of some truth information toperform the disambiguation, is used in the reconstruction. The left panel of Figure 5.1 showsthe ROC curve for the νµ event selection and the right panel shows the ROC curve for the νeevent selection. While this represents significant progress, the background rejection level hasnot yet reached the expectations set by the Fast MC studies described above. This is becausethe reconstruction is not sufficiently advanced to provide the expected discrimination; forexample, shower dE/dx currently provides relatively poor discrimination against neutralcurrent background for the νe CC event selection, even though this is expected to be apowerful discriminant. Work is ongoing to improve the reconstruction and re-evaluate theevent selection.
50
Figure 31: An example of an event where the merger of clusters has failed causing sectionsof track to be dropped. Left image shows the Ortho3D view, right image shows the wireplane display.
5.2 Oscillation Fits
Oscillation fitters are required to extracting measurements of physics parameters from theselected data sets. There are two oscillation fitters currently being developed for DUNE.VALOR, a neutrino fitting group first established in T2K, is able to perform fits to neardetector data that provide constraints for other oscillation fitters and is also able to performa joint fit to near and far detector data. The “Long-baseline Oscillation Analysis Fitter”(LOAF) fits far detector data and applies constraints based on separate near detector fitsand/or external data.
5.2.1 VALOR
The current version of the VALOR DUNE fit is an intermediate step towards the imple-mentation of the full joint ND-FD VALOR fit. It provides a constraint to the flux andcross-section systematic uncertainties, via the joint fit of neutrino flux and neutrino inter-action parameters to Near Detector event samples. Eighteen samples are used in the fit,corresponding to nine final state topologies for forward horn current (FHC) and reverse horncurrent (RHC) modes. These samples are νµ CC broken down into 1-track QE enhanced,2-track QE enhanced, 1π+, 1π0, 1π+ + 1π0, and other, wrong-sign νµ CC inclusive, νe CCinclusive, and NC inclusive. This set of final-state topologies is not entirely aligned withgenerator-level event classifications. To compare these measurements with theory, one needsto build an ND event rate model. This model is built from the convolution of flux, inter-action, and detector (acceptance) models, each of which have parameterized uncertainties.In VALOR, a simultaneous fit of neutrino flux and interaction parameters to ND samples isused to determine the values of these parameters, reduce their uncertainty, and obtain theircorrelations given the ND event rate constraint. In future revisions, the fully inclusive orother categories can be subdivided further to include neutrino-electron elastic scattering and
51
Figure 32: Analysis sample detection × selection efficiencies for the various signal and back-ground modes extracted from the Fast MC and used as inputs to GLoBES. Top: Used in theνe appearance sample. Bottom: Used in the νµ disappearance sample. Left: Neutrino beammode. Right: Antineutrino beam mode. The NC backgrounds (and νµ CC backgrounds forthe appearance mode) have been increased by a factor of 10 for visibility.
inverse muon decay events, as well as 3-track ∆-enhanced, coherent pion and other samplesthat can provide degeneracy resolution and improve the sensitivity of this analysis.
Measurements of a set of physics parameters in the presence of systematic parametersare obtained by comparing the observed and expected reconstructed kinematic distributionsfor a series of samples. Two dimensions (Ereco, yreco) are fit for the CC-like samples andone dimension (Evis) is fit for the NC-like samples. Here, Ereco is the reconstructed neutrinoenergy, yreco is the reconstructed inelasticity, and Evis is the reconstructed visible energy. Abinned likelihood-ratio method is typically used by VALOR. Most physics and systematicparameters in the VALOR fit come with prior constraints from external data; these con-straints are applied using a Gaussian penalty term. In the current VALOR analysis, whichprovides a constraint of the prior flux and cross-section systematic uncertainties, near detec-tor systematics are eliminated by profiling, as they are not correlated between the near andfar detectors.
MC templates are binned in the same kinematic variables that are used for fitting. Sep-arate templates are constructed for various true interaction modes; each template containsinformation on how the number of events, for each individual reconstructed bin, is distributedin a chosen space of true kinematic variables. In the current analysis, 52 MC templates cor-
52
Figure 33: ROC curve, showing background rejection vs signal efficiency, for the νµ CC eventselection (left) and the νe CC event selection (right).
Table 3: 52 Monte Carlo templates used by the VALOR fit to near detector data1. νµ CC QE 14. ν̄µ CC QE 27. νe CC QE 40. ν̄e CC QE2. νµ CC MEC 15. ν̄µ CC MEC 28. νe CC MEC 41. ν̄e CC MEC3. νµ CC 1π± 16. ν̄µ CC 1π± 29. νe CC 1π± 42. ν̄e CC 1π±
4. νµ CC 1π0 17. ν̄µ CC 1π0 30. νe CC 1π0 43. ν̄e CC 1π0
5. νµ CC 2π± 18. ν̄µ CC 2π± 31. νe CC 2π± 44. ν̄e CC 2π±
6. νµ CC 2π) 19. ν̄µ CC 2π0 32. νe CC 2π0 45. ν̄e CC 2π0
7. νµ CC 1π± + 1π0 20. ν̄µ CC 1π± + 1π0 33. νe CC 1π± + 1π0 46. ν̄e CC 1π± + 1π0
8. νµ CC coherent 21. ν̄µ CC coherent 34. νe CC coherent 47. ν̄e CC coherent9. νµ CC other 22. ν̄µ CC other 35. νe CC other 48. ν̄e CC other10. νµ NC 1π± 23. ν̄µ NC 1π± 36. νe NC 1π± 49. ν̄e NC 1π±
11. νµ NC 1π0 24. ν̄µ NC 1π0 37. νe NC 1π0 50. ν̄e NC 1π0
12. νµ NC coherent 25. ν̄µ NC coherent 38. νe NC coherent 51. ν̄e NC coherent13. νµ NC other 26. ν̄µ NC other 39. νe NC other 52. ν̄e NC other
responding to the true interaction modes shown in Table 3 are constructed for each of theeighteen fit samples. The nominal predictions are varied using the parameterized uncertain-ties of the flux, neutrino interaction and detector model. The VALOR DUNE fit provideswith best-fit values and correlations between all constrained parameters. Since the neardetector provides an event rate constraint, constrained flux and cross-section parametersbecome anti-correlated. The main output of the current analysis is a 156×156 covariancematrix for each proposed DUNE near detector design, which is used to provide constraintsfor the far detector fit. Work on a full fit to combined near detector and far detector data isongoing.
5.2.2 LOAF
The inputs to LOAF are far detector event spectra, flux and cross-section constraints from aVALOR fit to near detector data, and constraints on far-detector systematic uncertainties.The far detector spectra are taken either from a Fast Monte Carlo (Fast MC) or the fullLArSoft simulation and analysis. Currently Fast MC inputs are used because reconstruction
53
of the full Monte Carlo is still under development, but care has been taken to define acommon data format so that different types of simulation may be used interchangeably.
The LOAF fitter works by using prebuilt signal and background histograms combinedwith response functions which encode the information needed to alter those histograms givena particular set of fit parameter values. Since oscillation probabilities are a function of trueneutrino energy and species, template histograms of true neutrino energy broken out byspecies are required as inputs. Smearing functions are used to convert the oscillated trueneutrino energy spectra to reconstructed neutrino energy spectra. Systematics fluctuationsare applied either to the true energy spectrum before smearing, in the case of flux, cross-section, and nuclear models uncertainties, or to the reconstructed energy spectra, in the caseof reconstruction, efficiency, and other uncertainties from detector effects. These fluctuationsare applied via response functions which provide the relevant spectral distortions as functionsof systematics (fit) parameter changes. Parameter variations from nominal are used todetermine penalty terms in the fit χ2. The results of the VALOR ND fits are propagatedto the LOAF fitter via a covariance matrix that gives the covariance of the ensemble of fitparameters that describe the uncertainty on the predicted event rate at the FD. Detectorresponse systematics for the NDs have been marginalized and are not propagated. Theconstraints on the fit parameters are enforced through a penalty term calculated from thecovariance matrix. Additional fit parameters are used to allow the best-fit value of eachparameter to vary between samples. The statistical limit of the constraint on each parameteris estimated and used to set the allowed 1-sigma variation of that parameter in all othersamples. Each parameter is also allowed to vary based on theoretical limits on uncertaintyin the cross section ratios: νe/νµ, ντ/νµ, ν̄/ν, and NC/CC.
The LOAF fit χ2 assumes Poisson probability distributions for event counts and Gaussianprobability distributions for the priors used in penalty terms. In the case of correlated priorsa covariance matrix is used to determine the penalty term. The formulation is consistentwith the one used in the VALOR ND fits. The χ2 is minimized by the MIGRAD algorithmin MINUIT2 as implemented in ROOT or used by a Metropolis-Hasting algorithm in aMarkov Chain Monte Carlo (MCMC) method, allowing comparison of results between thetwo approaches. In the minimization approach, the minimum χ2 along with the set of best-fitparameters are reported.
For the purposes of a sensitivity study “mock data” must be generated. There are twoclasses of mock data used. The first is the most probable data, often referred to as the Asimovdata set. This is created by taking the “true” value of each parameter is the nominal valuefrom the simulation. This includes both systematic parameters and event rates. The secondclass of mock data are toy MC data sets in which the “true” values of each parameter arechosen at random based on the prior probability distributions. Again this applies for bothsystematic parameters and event rates, so takes into account statistical variations. In eithercase, since this is a hypothesis testing scheme, the parameter(s) of interest is set to a specificvalue which is different from than the nominal value, and not randomly generated. Asimovdata set studies can be performed with a single fit; however special care must be taken toensure correlations between data sets are properly considered. Toy MC data sets require aseries of fits and the results are given by examining the ensemble of fit results.
The LOAF fitting framework is currently working and producing results that seem rea-sonable. Work is ongoing to validate the results against the GLoBES-based sensitivity studies
54
presented in the DUNE CDR. Remaining work includes implementation of detector systemat-ics, implementation of uncertainty in the near/far flux ratio, implementation of correlationsbetween oscillation parameter priors, technical details of the interface with VALOR, andcomparisons between the MCMC and χ2 minimization techniques.
6 Detector Optimization Studies
6.1 Wire Spacing and Wire Angle Studies
LArTPCs provide excellent spatial and calorimetric resolutions for measuring the propertiesof neutrino interactions above a few MeV. It is important to optimize the TPC design inorder to maximize the physics sensitivities. In this section, we describe the Monte Carlo(MC) study of detector optimization, focusing on the impact of the TPC wire spacing andwire angle on the physics sensitivities.
The studies were done using the MC files produced in the MCC6.0 production [5]. Thedetails of the simulation and reconstruction software using in this MC production can befound in Ref [78].
6.1.1 Tracking Efficiency
The detection and identification (ID) of particles is a critical component of DUNE’s physicsprogram, as it impacts the ability of reconstruct neutrino events and other interactions. Inaddition, a vast number of ID algorithms relay on track reconstruction for a given particle.In this section we will describe tracking efficiencies for charged particles from neutrino beamevents. The simulation of such events follows the simulation chain; generation, its outputis passed to a geant4 based detector simulation in which particles are tracked when theytraverse in the LAr, the ionization electrons and scintillation photons are then digitized intoraw signal (ADC) in the process known as detector simulation. Full details of the simulationprocess can be found in Ref. [78].
The reconstruction of events in the TPC is done by a series of reconstruction algorithms.First is the signal processing, then Gaussian hit finder, disambiguation (induction wires arewrapped in the TPC design), cluster formation (2D objects) and finally 3D track (RS:) andvertex reconstruction performed by the Projection Matching Algorithm (PMA). Full detailsof the reconstruction process can be found in Ref. [78]. Tracking efficiency studies wereperform with Neutrino TrackingEff module.cc available at LArSoft/LArReco/ See [79].
Track reconstruction efficiency for a charged particle x± in e.g. neutrino beam interactionis defined as follows:
εx± =CC events with a x± particle and with a reconstructed track
CC events with a x± particle, (11)
where x± can be a µ±, π±, p, K+ and e−, the last two cases are from proton decay eventswhere p → K+ + νµ. To be more specific, the denominator includes events where a x±
particle was created and has deposited energy within any of the TPCs. The numeratorincludes events with a x± particle, and it has deposited energy within any of the TPCs witha reconstructed track associated to the x± particle. For cases where there is more than one
55
π± or protons, the leading particle will be used. The efficiency εx± can be calculated as afunction of energy, momentum, or angle. There are some cases when a charged particle hasmore than one track associated, which is caused by broken tracks. In order to choose thebest track a metric called “completeness” is defined as
Completeness =track hits∑
i
(hit energy on track from a particle)itotal hits energy from a particle
(12)
Another useful metric is track purity, which estimates the level of “hit contamination” fromanother particle; a metric called “purity” was defined as
Purity =track hits∑
i
(hit energy on track from a particle)itotal hits energy on the tack
(13)
To calculate tracking efficiencies when there is more than one track associated to thesame particle, the best track is chosen by choosing the track with the highest Completeness.
Track Completeness0 0.2 0.4 0.6 0.8 1 1.2
0200400600800
10001200140016001800200022002400
h_Ecomplet_leptonEntries 9080Mean 0.8433RMS 0.1993
Ecomplet Lepton
Track Purity0 0.2 0.4 0.6 0.8 1 1.2
0
200400
600800
1000
120014001600
180020002200
h_Efrac_leptonEntries 9080Mean 0.8483RMS 0.04559
Efrac Lepton
Muon Momentum (GeV)0 1 2 3 4 5
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
True length - Reco length (cm)100− 80− 60− 40− 20− 0 20 40 60 80 1000
500
1000
1500
2000
2500
3000
3500 h_trackRes_leptonEntries 9080Mean 6.728RMS 15.09
Muon Residual
Figure 34: Track completeness (top left). Track purity (top right). Tracking efficiency(bottom left). Tracking residual (bottom right).
To test the tracking reconstruction in a isotropic way, we calculate the tracking efficiencyfor µ− with uniform momentum distribution from 0.1 to 5.0 GeV, generated isotropicallyaround the detector. The differences with a neutrino beam sample is that in the neutrinobeam sample, most of the particles will travel forward due to the boost from the incoming
56
neutrino so the phase-space is reduced to mostly forward interactions. By using an isotropicsample we can test the reconstruction in all directions. Fig. 34 shows Completeness, Purity,tracking efficiency and track residual for µ− using the particle cannon sample. The trackresidual is calculated using the particle length using only the trajectory points of the particlethat are inside of a TPC (true length) and the reconstructed track length (reco length). thesecond peak in the residual is due to broken tracks either by delta rays or by the muontraversing through different TPCs. Looking at track completeness there are two main factorsthat can cause track completeness < 1: there are some cases where a muon traverses differentTPCs and its track sometimes is not stitched properly so that it becomes two tracks (brokentrack); also delta rays can cause a track to be split. Both cases affect the residual and theseexamples appear in the residual as a second peak. In the case of track purity the peak isaround ∼0.84; for most of these muon tracks there is 16% EM activity along the track, i.e.delta rays or Bremsstrahlung photons. Overall the tracking efficiency is 95.4%.
Figure 35: Event display for a CC νµ event in LAr (collection, induction1 and induction2planes).
Figure 35 shows a fully reconstructed event produced in a CC νµ interaction in LAr.There are three panels (from top to bottom) collection plane, induction1 and induction2plane. Figure 36 shows tracking efficiencies for µ− produced in a CC νµ interaction in LAr.Tracking efficiencies are pretty much identical for beam and atmospheric events as well forneutrino and antineutrinos. In this document we will discussed only efficiencies for CC νµinteraction in LAr. Overall the tracking efficiency for muons is 98.2%, which is quite aremarkable performance from the pmttrack module. An interesting feature is that trackingefficiency drops at ∼ 70 degrees. This is mainly due to particles traveling almost parallel tothe collection wires, causing one of the views to be missing in the reconstruction.
Figs. 37, 38 and 39 show π−, π+ and p tracking efficiencies and residuals respectively.In all cases the tracking efficiency drops at low momentum, more specifically, for protons at∼70 MeV of kinetic energy and for pions at ∼ 100 MeV of kinetic energy. Due to the factthat LAr is a dense material, hadronic interactions make it more challenging to reconstruct
57
Neutrino Energy (GeV)0 5 10 15 20 25
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
Muon Momentum (GeV)0 5 10 15 20 25
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
Theta w.r.t beam direction (Degrees)0 10 20 30 40 50 60 70 80 90
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
True Length - Reco Length (cm)100− 80− 60− 40− 20− 0 20 40 60 80 1000
1000
2000
3000
4000
5000
6000
7000
8000
h_trackRes_leptonEntries 41216Mean 6.189RMS 17.83
Muon Residual
Figure 36: Tracking Efficiency for µ− produced in a CC νµ interaction in LAr. (top left)as a function of incoming neutrino energy. (top right) as a function of muon momentum,(bottom left) as a function of the muon angle w.r.t. the incoming neutrino. (bottom right)track length residual
58
Pion Momentum (GeV)0 0.5 1 1.5 2 2.5 3
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
True Length - Reco Length (cm)100− 80− 60− 40− 20− 0 20 40 60 80 1000
100
200
300
400
500
600
h_trackRes_pion_minus
Entries 5692Mean 4.971RMS 19.52
Pion - Residual
Figure 37: Tracking Efficiency of the leading π− produced in a CC νµ interaction in LAr.(left) as a function of pion momentum, (right) track length residual
Pion Momentum (GeV)0 0.5 1 1.5 2 2.5 3
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
True Length - Reco Length (cm)100− 80− 60− 40− 20− 0 20 40 60 80 1000
200
400
600
800
1000
1200
1400
16001800
2000
2200
h_trackRes_pion_plus
Entries 16021Mean 2.866RMS 18.71
Pion + Residual
Figure 38: Tracking Efficiency of the leading π+ produced in a CC νµ interaction in LAr.(left) as a function of pion momentum, (right) track length residual
the tracks for low momentum particles before they undergo a hadronic interaction.Next we present tracking efficiency using various detector configurations with different
wire spacing, wire angles and TPC orientations. What we call Default geometry implies5 mm wire pitch and 36 degree wire angle, 3mm is 3 mm wire pitch, 45Deg is 45 degree wireangle, and r90 Deg beam is a 90 degree rotation of the beam along the z-axis, mimicking aTPC rotation.
Results from the different geometries with different wire pitch and different angle arepretty similar. This can be understood because trajectory fits (tracks) use many data points,over much larger scales than wire pitch. For the case of r90Deg beam the drop in efficiencyis due to low hit finding efficiency in this direction. A summary of tracking efficiencies canbe found in Table 4. Figure 40 shows tracking efficiencies for the different geometries forµ−, π−, π+ and p as a function of the true length. From the plots there is a 2.5 cm thresholdfor a 3D track reconstruction regardless of the particle. Overall, the 3mm configuration gives
59
Proton Momentum (GeV)0 0.5 1 1.5 2 2.5 3
Tra
ckin
g E
ffici
ency
0
0.2
0.4
0.6
0.8
1
True Length - Reco Length (cm)100− 80− 60− 40− 20− 0 20 40 60 80 1000
1000
2000
3000
4000
5000
h_trackRes_protonEntries 21226Mean 0.1822− RMS 12.88
Proton Residual
Figure 39: Tracking Efficiency of the leading p produced in a CC νµ interaction in LAr.(left) as a function of proton momentum, (right) track length residual
Muon True Length (cm)0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmr90Deg beamDefault
True Length (cm)(+)Pion0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmr90Deg beamDefault
True Length (cm)(-)Pion0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmr90Deg beamDefault
Proton True Length (cm)0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmr90Deg beamDefault
Figure 40: Tracking efficiency for µ− and leading π+, π− and p produced in a CC νµinteraction in LAr as a true length.
the better results.An important topic in DUNE’s physics program is the search for proton decay. The
60
Table 4: Tracking Efficiencies for CC νµ events (statistical error only)Default (%) 3mm (%) 45Deg (%) r90Deg beam (%)
Muon 98.2 ± 0.06 98.7 ± 0.06 98.0 ± 0.07 72.2 ± 0.21Proton 61.0 ± 0.26 64.0 ± 0.26 60.8 ± 0.26 49.5 ± 0.26Pion+ 65.8 ± 0.30 67.8 ± 0.31 67.5 ± 0.30 50.1 ± 0.31Pion− 61.8 ± 0.50 62.0 ± 0.52 62.2 ± 0.51 42.6 ± 0.51
detection of proton decay would represent experimental evidence for GUTs. With the world’slargest particle detector deep underground, DUNE is the ideal detector to search for sucha process. Here we present tracking studies of the so called ”golden channel” p → K+ + ν̄.Thanks to the TPC design it’s possible to reconstruct the products of the kaon decay,in specific for the mode K+ → µ+ + νµ and also the muon decay µ+ → e+ + νe + ν̄µ.Overall the tracking efficiency for kaons is 60.0%, and 68.3% for muon coming from the kaondecay and 63.1% for Michel electrons coming from the muon decay. Figure 41 shows a fullyreconstructed proton decay event.
Another study was performed by looking at different detector configurations where twodifferent reconstructions were applied: TrajCluster and LineCluster. We look at the resultsas efficiencies for K+, µ+, e+. Overall there is an increase on the efficiency of 2% when usingTrajCluster for K+ and µ+, and a 9.0% increase on the efficiency for the Michel electrone+. In summary by using TrajCluster we can lower the threshold for 3D tracks to below2.5 cm specifically for Michel electrons. Figure 42 shows the tracking efficiency for K+, µ+
and Michel electrons as a function of true length for the different detector configurations.
Figure 41: Event display for p→ K++ ν̄ event in LAr (collection, induction1 and induction2planes).
61
Kaon True Length (cm)0 5 10 15 20 25 30 35 40 45
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmDefault
Kaon True Length (cm)0 5 10 15 20 25 30 35 40 45
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmDefault
Muon True Length (cm)0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmDefault
Muon True Length (cm)0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmDefault
Michel Electron True Length (cm)0 10 20 30 40 50
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmDefault
Michel Electron True Length (cm)0 10 20 30 40 50
0
0.2
0.4
0.6
0.8
1
1.2
1.4 45Deg3mmDefault
Figure 42: Tracking efficiency K+, µ+ and Michel electron (left is with LineClusters, rightTrajCluster).
6.1.2 e/γ Separation
Neutrino oscillation analysis is based on the identification of the incident neutrino flavor andestimation of neutrino energy. Electron neutrino charged current (νeCC) interactions are tobe found among charged and neutral current interactions of other flavors of neutrinos in thefar detector data. Several features of the reconstructed events may be used as signatures ofνeCC. In this chapter we describe potential of reconstruction of two such features which are
62
expected to be strongly discriminating between νeCC (signal) events and background events,namely the ionization density in the initial part of the electron candidate cascade (dE/dx)and detection of the displacement between the starting point of the electron candidate cas-cade and neutrino interaction vertex (gap). These two features should allow us to distinguishthe single electron appearing in the νeCC interaction (single m.i.p. dE/dx, no gap detected)from gamma conversion induced cascades (double m.i.p. dE/dx, possible detection of thegap) originating from the decays of π0’s appearing in CC and NC interactions of neutrinosof any flavor. The gap and dE/dx may be reconstructed with different efficiency in variouscombinations of the detector design parameters. In this section we do not discuss νeCC se-lection inefficiency related to physics, e.g. energy dependent fraction of photons undergoingCompton scattering (thus producing single electron signal-like cascades).
Full event automatic reconstruction is under development. Most components of trackreconstruction and particle hierarchy are in place but complete and efficient electromagneticshower (EM shower) reconstruction and detailed pattern recognition of vertex region arestill being developed. Therefore we used the combination of simulation and reconstructionto test potential combinations of detector parameters.
There are five pieces of reconstruction required for νe selection and related DUNE physics:
• Recognition of a shower-like object pointing to the neutrino interaction vertex.
• Identification of a visible starting point of the shower (point where the cascade startsto be separated from other tracks near the vertex region).
• Cascade direction reconstruction.
• Electron / photon identification.
• Shower energy reconstruction.
Fiducial volume cuts and corrections for non-containment which are required for theenergy measurement have not been studied.
Pattern recognition may handle under some conditions showers which cross dead spaces.The minimum reasonable requirement to identify a shower is to have the initial track-likepart and the showering part being visible within the same TPC volume. The radiation lengthof 14 cm is in the agreement with the observation of about 20 cm along the shower axis thatis needed for good reconstruction results. Taking into account the distribution of the anglebetween the electron and neutrino direction (less than 10% of events above 40 degrees andpractically no events above 60 degrees) the first guess for the reasonable fiducial volume cutis about 10-15 cm on lateral walls. If the cathode thickness is small (less than 1 cm), onecan consider no cut on the cathode side to be required by the pattern recognition. On theother hand, any field distortions due to supporting structures may result in a significant cut.
Several tracks and showers may overlap in 2D projections near the vertex of neutrinoevents. The shower part useful for the reconstruction (hits of the shower separated from othertracks) can be displaced from the vertex according to the distributions shown in Figure 43.Significant hints from MC truth were used in order to show in the plot the expected upperlimit of the reconstruction precision.
63
Figure 43: Distance of the visible shower starting point from neutrino interaction vertex.MC truth information (neutrino vertex position and electron direction) is used to select 2Dclusters which are used to reconstruct 3D shower direction in order to approximate bestpossible reconstruction result (3mm wire pitch was used here).
The present method can select as long of an initial section of the shower as useful for thedirection reconstruction. The selected part consists of a clean electron (pair) track and theshowering part that is not yet strongly developed in the transverse directions. Requirementsfor this technique are similar to the identification of the shower-like object: length comparableto the radiation length is preferable, although even 2 cm sections are useful in the directionreconstruction in case of early developing showers.
Current pattern recognition algorithms are not precise enough in selection of electroncandidate cascades in neutrino event topologies expected for DUNE. Especially problematicis the assignment of hits to particles near the primary vertex. The optimization study needsto evaluate the electron/gamma selection performance that is potentially possible to achieve,regardless of the present state-of-the-art in pattern recognition.
Wire pitch may have an impact on the particle identification which is based on thedE/dx calculation on small scale distances. On the contrary, particle trajectory and vertexreconstruction are based on features much larger than wire pitch and are not sensitive to thedifference between wire pitch options. Wire angle between different planes can have someimpact on particle identification since the plane selected for the calculations may be chosenfrom a more or less wide range of available orientations.
These parameters were assumed in the simulation: 3 ms electron lifetime, longitudinaland transverse diffusion, simplified E-field impulse response (signal induced on the singlewire) and relatively high S/N. Both low S/N ratio and low electron lifetime affect efficiencyof zero suppression, deconvolution and hit finding algorithms. All of them affect the fullreconstruction chain and particle identification efficiencies.
Capabilities of pattern recognition algorithms at the present stage are not sufficient toselect EM showers in the crowded region of the neutrino vertex. Reconstruction tests usingfull neutrino events are not yet sensitive to the difference in the detector parameters. Forthis reason studies were performed in the following conditions:
• Single, isolated electrons and photons were simulated with the uniform distribution in
64
the far detector workspace geometry, with the flat energy distribution [0.1, 5] GeV.
• Wire signal deconvolution and hit reconstruction were applied as in the standard re-construction chain.
• Plane with the longest projection of the reconstructed initial segment of the cascadewas used for dE/dx calculations.
• dE/dx along the reconstructed direction of cascade was calculated using PMA (PMAl-gTrackMaker module).
• Effect of hadron tracks overlapping with the cascade in the vertex region was simulatedby removing hits up to some range from the starting point according to Figure 43 andby smearing the calculated distance of hits to the starting point in photon cascades,also according to Figure 43.
cosine0.99 0.991 0.992 0.993 0.994 0.995 0.996 0.997 0.998 0.999 10
200
400
600
800
1000
1200
1400
1600
1800
2000Default, electrons
Default, photons
Figure 44: Reconstruction of cascade’s initial direction for electrons and photons in thedefault configuration.
The above conditions allow us to mimic the real complexity of the vertex region and sep-arate inefficiencies due to t the current pattern recognition status. Direction reconstructionis limited by the properties of EM showers (effects of scattering and showering of electrons,two tracks from the vertex in case of photon conversion) and orientation of the shower withrespect to the readout planes. Figure 44 shows the angle of reconstructed direction with re-spect to the truth value. As expected resolution is worse for photons which convert to e+e-pairs. However different wire pitch/orientation configurations do not affect reconstructionas shown on Figure 45.
Studies of shower direction reconstruction show also that the upper limit of the angularresolution is about 6 degrees on average for the presented sample, mainly due to largescattering of low energy electrons and showering in higher energies. At the present stageof shower direction reconstruction the above limit is achievable at the minimal angle of 10degrees with respect to the wire planes.
65
cosine0.99 0.991 0.992 0.993 0.994 0.995 0.996 0.997 0.998 0.999 10
200
400
600
800
1000
1200
1400
1600
1800
2000
2200 Default, electrons
3mm, electrons
45deg, electrons
cosine0.99 0.991 0.992 0.993 0.994 0.995 0.996 0.997 0.998 0.999 10
100
200
300
400
500
600
700
800
900
Default, photons
3mm, photons
45deg, photons
Figure 45: Reconstruction of the cascade’s initial direction for electrons (left) and photons(right) for three detector configuration options.
dE/dx [MeV/cm]0 2 4 6 8 10 12 14
0
0.05
0.1
0.15
0.2
0.25
0.3Default, electrons
3mm, electrons
45deg, electrons
Default, photons
3mm, photons
45deg, photons
Figure 46: Averaged dE/dx from the initial part of cascade for electrons and photons,calculated for different wire configurations.
66
The classic method of discrimination between electron and photon induced cascades isbased on dE/dx averaged over the initial, track-like section of the cascade. In the presentedstudies the length of the initial section was optimized to obtain best separation of simulatedcascades. In addition we have found that electron selection efficiency of the method canbe increased if the first part separated from other tracks in the vertex is used instead of apart up to a fixed distance from the vertex (events overlapped at a longer distances fromthe neutrino interaction are not lost). The best separation was obtained for a short (1.2-1.5 cm) length for the cascade section. Figure 46 shows distribution of dE/dx for variouswire pitch/orientations.
range [cm]1 2 3 4 5 6 7 8 9 10
dE
/dx [
Me
V/c
m]
0
1
2
3
4
5
6
7
electron
photon
Figure 47: Left: dE/dx as a function of range for electrons (blue) and photons (red). Right:color shows the probability fit that given (range; dE/dx) point belongs to electron cascade;final probability value that a cascade is due to electron or photon conversion is calculated asa product of probabilities of (range; dE/dx) points selected in the initial section of cascade.
An alternative method for electron and photon discrimination was proposed. It allowsus to exploit in more detail the statistical differences in dE/dx evolution as a function of therange from the vertex, and dE/dx fluctuations (short increase of ionization due to delta rays)are handled better. The probability that a given shower is initiated by electron or photonis computed instead of using the average dE/dx as a discriminating value; illustration forthe default configuration is shown in Figure 47. Electron and photon dE/dx profile patternswere prepared for the range up to about 15 cm from the primary vertex. Similarly to theprevious approach, dE/dx points in the first section separated from other tracks in the eventare used (usually 2, 3 hits). A probability that the data point belongs to gamma, electronor to ”none” class is computed for each point. Finally, probabilities calculated for all pointsare accumulated and form the discriminating value. The final signal selection efficiency plotas a function of background rejection is shown in Figure 48. For an example value of 90%electron selection efficiency, background reduction by a factor of 25% can be obtained forthe 3 mm wire pitch or 45 degree wire orientation options.
67
Background rejection [%]50 60 70 80 90 100
Sig
na
l se
lectio
n e
ffic
ien
cy [
%]
0
20
40
60
80
100
Default
3mm
45deg
Background rejection [%]80 85 90 95 100
Sig
na
l se
lectio
n e
ffic
ien
cy [
%]
80
82
84
86
88
90
92
94
96
98
100
Default
3mm
45deg
Figure 48: Electron (signal) selection efficiency as a function of photon (background) re-jection for three detector configurations. Right plot shows enlarged region from the leftplot to show differences in signal efficiency in the interesting range of signal selection andbackground rejection.
6.2 Perpendicular vs. Parallel APA Comparison
LArTPC is a new detector technology under rapid development for neutrino physics. Thepresent challenges include i) high voltage, ii) high argon purity, iii) low electronic noise, iv)negligible ”dead” channels, v) proper TPC signal processing, and vi) event reconstruction.Once the challenges regarding the instrumentation performance, quality control, and inte-gration are overcome, LArTPC event reconstruction is likely to be one of the main remainingbottlenecks to extract physics for experiments. Therefore, it is crucial to examine hardwareoptions which can reduce the challenges in LArTPC event reconstruction. Out of manypotential detector parameters to optimize, the orientation of electric field of the TPC withrespect to the neutrino beam is the most important one. We note that the drift electric fieldof the TPC defines the symmetry axis of the device. For isotropic events such as protondecay or astrophysical neutrinos the orientation of this axis with respect to the events cannotbe specified. But for accelerator beam events, this choice has to be justified on the basisof the most important requirements of event reconstruction, vertex, and energy resolution.The TPC electric field can be either parallel (APA perpendicular) or perpendicular (APAparallel) to the beam. All other orientations are perturbations on these two fundamentalchoices. A technote [80] has been prepared to systematically analyze this problem based onthe tools available. In this brief note, we summarize the most important observations fromthis study.
Table 5 summarizes the comparison of the two detector orientations. It should be recalledthat there are several detailed requirements on the detector performance for beam neutrinophysics. It is required that most (>80%) of the charged current electron neutrino events bereconstructed with excellent energy resolution. The reconstruction must identify the vertexof the events with ∼ 1 cm resolution in all dimensions. The neutral current background
68
to charged current electron neutrino events is to be suppressed using gap and dE/dx (e/γseparation) cuts so that the contamination is < few %.
Figure 49: 3D images are reconstructed for a simulated 3-GeV νe charge-current interaction.The left (right) panel shows the reconstructed 3D images with the perpendicular (parallel)TPC configuration.
As shown in Fig. 49, the main advantage of a perpendicular APA design is in the eventreconstruction, which would result in a higher signal selection efficiency, lower backgroundcontamination, and better angular resolution. The angular resolution is crucial for some par-ticular event topologies, such as quasi-elastic scattering and neutrino-electron elastic scat-tering. We have performed studies based on fast Monte Carlo augmented with appropriatedetector geometry simulations. We focus initially on the topology in which particles tracksare near parallel to the wire planes (APA) near the vertex. Such tracks cannot be recon-structed in 3D a priori because the signals from entire trajectory are at the same time dueto the finite longitudinal (time) resolution. For the APA parallel geometry: we find thata large fraction (∼ 30%) of the charged current events will have the lepton in the parallelconfiguration when the APA are parallel to the beam. These events are likely to have poorangular and energy resolutions depending on the characteristics of the events (quasi-elastics,resonance, DIS, etc.). Similar number of events will have hadronic particles or photons par-allel to the APA. Furthermore, we calculate that ∼ 10% of the entire charged current samplewill have overlaps of leptonic and hadronic particles in the parallel plane, which is likely themost difficult to be reconstructed and correctly identified. We also find that the number ofneutral current π0 background events with overlapping hadrons and converting electrons inthe parallel plane is high, about 10% compared with the entire charge current signal sample.There is an additional ∼ 20% background where a charged pion overlaps with a π0-decaygamma which converts to an electron further away from the vertex in the parallel plane.
69
Such a background could mimic a νe signal with a similar dE/dx distribution. The effect ofthese topologies on vertex, energy, and particle identification is under further evaluation. Wefind that these problematic topologies are considerably reduced in the APA perpendiculargeometry.
Furthermore, as shown in Table. 6, the position resolution along the drift field direction(longitudinal) is better than the transverse directions parallel to the wire plane. Therefore,a better e±/γ separation is expected for the perpendicular APA design, which can reducethe background level. In addition, a 30% cost reduction on the electronics is expected aslisted in Table. 5.
The main disadvantage of the perpendicular APA orientation is the reduction of thesignal to noise ratio in the induction plane. It has been shown that this should not be aproblem if the expected noise level on the cold electronics are achieved on DUNE. However,perpendicular APA orientation will suffer more if there is a sizable increase on the noise leveldue to excess noise. Therefore, it is important to demonstrate the expected noise level canbe achieved in the future prototype detectors. 7
Table 5: Summary of advantages and disadvantages of perpendicular APA design with re-spect to the default parallel APA design. The text in blue represents the major advantages.The text in red represents the main disadvantages. The text in black is expected to havenegligible impacts on the physics.
Pros ConsPattern recognition More information in imaging. 8 Busy regions due to wrapped
Better e/γ separation. wire. 9
Charge response Less variations in the induction Loss in induction chargefield response. resolution. 10
Position resolution Better angular resolution for Worse angular resolution forlong tracks. non-parallel short tracks. 11
Less sensitive to the actualwire location.
Energy resolution Better dE/dx calculation Energy resolution due to APAfor quenching correction. 12 gaps. 13
Less capabilities in momentumdetermination from multiplescattering. 14
Other considerations Save ∼ 30% Channels. 15
Electronics design specificationis suitable.
7 LARIAT has demonstrated that the expected noise level can be achieved. MicroBooNE has demon-strated that the expected noise level can be achieved after a software noise filter [81]. Hardware upgrades areon-going to achieve the expected noise level before software noise filter. More unique features in DUNE, suchas the mixed analog-digital front-end and the modular design, will be tested in the protoDUNE detector.
8The improvements in the pattern recognition leads to improvement in almost all aspects of the laterevent reconstruction chain including energy and angular calculation.
70
Table 6: Comparison of the resolution between the longitudinal and transverse directions forLArTPC operating at 500 V/cm electric field. The resolution in the longitudinal dimensionis expected to be much better than that of the transverse dimension due to better resolutionand finer digitization.
Property Longitudinal Dimension Transverse DimensionDigitization length 0.8 mm 5 mm 16
Diffusion σ 17 < 1.7 mm < 2.4 mmElectronics shaping σ 1.3 mm 18 N/AField response function σ ∼1.1 mm 5 mm 19
Overall resolution < 2.5 mm > 5 mm
6.3 Considerations of implementing an additional readout wireplane in the single-phase LArTPC
In this section, we summarize the considerations of implementing an additional inductionwire plane in the single-phase LArTPC to enhance its performance.
In the current design of the DUNE single-phase LArTPC, there are three readout wireplanes. In front of these readout planes, there is a grid wire plane added in order to shape theimpulse field response function. On the back of these readout planes, there is a wire plane toprevent the collection of ionization charge generated inside the anode plane assembly (APA).Bias voltage is applied on these (five) wire planes, so that the ionization electrons will driftat 100% transparency through the grid plane and first two readout wire planes before allcollecting on the last readout plane. The first two readout wire planes are commonly referredto as the induction plane and the last readout wire plane is referred to as the collection plane.The impulse field response function has a bipolar and unipolar shape for the induction andcollection planes, respectively.
There are three reasons that adding one more induction wire plane will significantlyincrease the performance of the single-phase LArTPC. They are:
• Ambiguities reduction,
9The proposal of the single-sided APA will completely solve this issue.10At expected electronics level of DUNE, the impact of increment in the charge resolution from induction
wire plane on the induction plane hit selection efficiency is small for the νe signal. Nevertheless, if there isa sizable increase of the electronic noise, perpendicular APA configuration will suffer more.
11For 1 GeV/c muon, the short track is defined as the traveling length shorter than 20 cm.12The improvement of dE/dx is coming from improvement of the position resolution as well as pattern
recognition.13The leading contributor is shown to be neutrons generated in the neutrino and subsequent hadronic
interactions.14Due to the vast size of DUNE detector, the impact of this loss of capability on νµ charge-current
interaction is expected to be small.15There is a 3% reduction on the fiducial volume, which is changed to +0.9% gain with 3.8 meter drift
distance.16This length is basically the wire pitch.17Maximum drift distance is assumed to be 3.6 m.18The shaping time is assumed to be 2 µs.19The resolution due to induction is expected to be comparable to the wire pitch.
71
• More resistance to dead channels,
• Increase the acceptance for the induction signal through robust ROI finding.
In the following, we will briefly describe each reason.In single-phase LArTPC, the 1D wire-based readout is necessary due to the financial
constraint on the total number of channels that can be afforded and the physical constrainton the total electronics consumption power that can be allowed in the LAr. Compared tothe 2D pad-based readout, the (multiple) 1D wire-based readout significantly reduces thetotal number of channels at the cost of increases in ambiguities. Figure 50 shows the numberof reconstructed potential hits as a function of the number of real hits in a fixed time slice.A reconstructed potential hit is defined as a hit where multiple hit wires (one from eachreadout plane) cross. These wires may not be necessarily originated from the same real hit.Therefore the number of reconstructed potential hits is larger than the number of real hits.The different colored solid curves represent different number of total readout wire planesfrom 2 to 6. It is clear that more wire planes would reduce the ambiguities. For DUNE (3wire plane case), the number of potential hits roughly ranges from one to fifty for νe charged-current interactions. Therefore, increasing the number of readout wire planes from three tofour represents a significant step towards reducing the ambiguities, which is crucial for clearlyidentifying activity near the neutrino interaction vertex (one of the busiest regions).
Figure 50: Illustration of ambiguities with the number of reconstructed potential hits as afunction of the number of real hits in a 1D wire-based readout detector. Different coloredcurves represent different number of wire planes from 2 (black) to 6. The no-ambiguitysituation where the number of reconstructed potential hits is exactly the same as the numberof real hits is labeled as “real == potential”. The solid curves represent the situation whenall the channels are functioning. The dashed curves represent the situation where some ofthe channels are dead. The left (right) panel represents the situation where 1% (5%) of thechannels are dead. See the text for more discussion.
In practice, it is unrealistic to have 100% working channels for a 10kt detector. Forexample, in the current-generation large LArTPC experiments, the ratio between the numberof dead channels and the number of total channels is above 10%. In the next-generation largeLArTPCs, we expect this ratio to be significantly reduced. Nevertheless, it is unrealistic to
72
assume zero dead channels. Assuming “p” is the efficiency of a single wire plane, the volumeefficiency can thus be estimated as εn = pn given “n” number of wire planes. This volumeefficiency can be enhanced if fewer planes are required in the reconstruction. For example,if we require one less readout plane to be used in the reconstruction, the volume efficiencyis εn−1 = pn + n · (1 − p) · pn−1, which can be much higher than εn. However, the increasein the volume efficiency is at the cost of increasing the number of ambiguities. AssumingFn represents the number of reconstructed potential hits given “n” number of wire planes,the number of reconstructed potential hits using one less plane in the reconstruction can beestimated as (Fn+(1−p) ·n · (Fn−1−Fn)) · εn−1. Basically, when one less readout wire planeis required in the reconstruction, the number of reconstructed potential hits Fn will receivea leakage from that of the “n-1” wire plane Fn−1. The left (right) panel of Fig. 50 illustratesthe situation for 1% (5%) dead channels. Since the ambiguity with three wire planes is muchreduced compared to that of the two wire planes, the four wire planes are much more robustagainst the potential dead channels compared to the three wire planes case.
Besides the existence of ambiguity due to the 1D wire-based readout, another big chal-lenge of the single-phase LArTPC is the processing of the induction wire plane signal. Whenthe impulse field response of the collection wire plane is unipolar, the impulse field responseof the induction wire plane is bipolar. The integration of the induction impulse responsefunction over time is close to zero, as none of the ionization electron is collected by the in-duction plane. Depending on the original ionization electron distribution, the positive lobeof field response could cancel with the negative lobe leading to overall smaller signal heightscompared to those from collection wire plane. Therefore, the induction wire plane signal ismuch more sensitive to noise compared to the collection wire plane signal. The implementa-tion of the cold electronics that significantly reduces the electronic noise as shown in Ref. [81]represents a major step towards improving the induction wire plane signal. Furthermore,the implementation of the 2D deconvolution in the signal processing chain [82] representsanother significant step in improving the induction wire plane signal. Despite these efforts,there is still a limitation in the induction plane signal, as the signal to noise ratio becomessmaller with the increase of the signal (time) length. In order to maximize the signal to noiseratio in the induction plane, the region-of-interest (ROI) technique is crucial. A ROI is aregion that just contains the signal, so that the electronic noise, especially the low-frequencyone, is minimized during the signal processing. In practice, while finding the ROI is a relativesimple task for a short signal, it is much more difficult to reliably find the ROI for a longsignal due to the smaller signal to noise ratio. Adding a new readout wire plane can signifi-cantly improve the ROI finding. Recall that a real ionization signal would generate currentson each wire plane simultaneously 20. Therefore, a long ROI can be built from short ROIsfrom the other wire planes requiring the wires from different planes crossing at the samepoint. This technique works for tracks that are not traveling completely perpendicular toa wire plane. In principle, this technique can work for the sitution of three readout wireplanes. In this case, the ROI has to be built from the other two wire planes. However, asshown in Fig. 50, the number of ambiguities for three wire planes is much reduced than thatof two wire planes. Therefore, the aforementioned robust ROI finding technique is much less
20The coincidence is after taking into account the short time needed for ionization electrons traveling fromone wire plane to the next.
73
powerful in the three readout wire planes case. Adding one more induction wire plane cansignificantly improve the situation.
Technically, in the current single-phase APA design, the grid plane, which is at the correctbias voltage, can be converted into the fourth induction wire plane once it is read out.Ref. [82] shows that the signal processing of the first induction wire plane in MicroBooNE(without a grid plane) is successful. Therefore, the original requirement of shaping theinduction impulse field response with a grid plane can be relaxed. The angle of the gridplane wire will need to be changed so that there is no degeneracy in the wire orientationamong all readout planes.
Based on these studies, we also emphasize it is important to consider three or morereadout planes for the dual-phase LArTPC design. Although the dual-phase design is freeof the complication from induction wire planes, the considerations of reducing ambiguitiesdue to 1D wire-based readout still apply.
6.4 Photon Detector System Task Force
The photon detector system task force was formed in August 2016 to address two questions:do the existing photon detector (PD) requirements meet the DUNE physics goals, and doimprovements to the DUNE physics program warrant the costs of increasing the performanceof the PD system? In particular, the charge to the PD task force is defined as determiningwhether:
1. The existing PDS and TPC requirements are sufficient define a detector capable of per-forming scientifically relevant measurements that support the DUNE physics program,which for this charge shall be defined as both the beam physics program and non-beamphysics associated with proton decay, atmospheric neutrinos, and SNB events.
2. The existing light detection response parameter of 0.1 photoelectron-per-MeV of en-ergy deposition is sufficient to support the DUNE physics program, and how rapidlyscientific benefits would increase as this light yield is increased.
3. Addition of specific mean detection efficiency requirements for various minimum visibleenergy depositions would significantly enhance the DUNE physics program.
4. Addition of specific uniformity of detection efficiency requirements for various minimumvisible energy depositions would significantly enhance the DUNE physics program.
The highest level (L2) Requirements for the PD system can be found in the LBNF-DUNErequirements document [83].
The PD task force will provide a report addressing these charges by March 2017 incoordination with the full Far Detector Task Force. In order to achieve this goal the PDtask force has begun establishing a team that includes membership from the PD workinggroup, as well as the supernova, nucleon decay, and atmospheric neutrino physics groups.Preliminary work by the task force will involve reproducing existing PD-related physicsstudies using the more sophisticated tools developed by the DUNE collaboration. Followingthis effort, members of the task force will work to implement enhanced PD performance into
74
the DUNE far detector simulations in order to gauge the level of physics performance thatcan be achieved through such improvements.
A PD task force workshop will be held in January in order to assess the progress of thegroup and to prioritize the task force effort going forward. Table 7 outlines the task forcetimeline.
Milestone Target Date
First task force meeting Sep (2016)Weekly meeting begin Oct (2016)Reproduction of existing studies complete Nov (2016)PD Task Force Workshop Jan (2017)Finalizing Task Force Report Mar (2017)Deliver Task Force Report Mar (2017)
Table 7: Photon detector task force timeline.
7 Summary and Next Steps
Great progress has been made toward the task force goals in the past year. This sectionsummarizes the accomplishments and next steps for each item in the charge.
FD Simulation and Reconstruction Chain: The full simulation and reconstructionchain is functioning. The simulation and reconstruction working group regularly produceslarge sets of MC for analysis. Simulation and reconstruction algorithms will continue to berefined, but the infrastructure is in place.
Simulation and Reconstruction for SNB and Nucleon Decay Physics: Generatorsfor supernova neutrinos, atmospheric neutrinos, cosmic muons, and nucleon decay physicshave all been successfully incorporated into LArSoft, and samples have been generated usingthe DUNE geometry. Reconstruction performance is being tested for nucleon decay physics,and reconstruction for low-energy neutrinos is starting to be developed. The next stepswill be refining the reconstruction and examining physics performance for different detectorgeometries.
Long-Baseline Sensitivity Update: Event selection algorithms have been developed us-ing beam MC, and oscillation fitters are being validated. In the near future, the first sen-sitivities using full simulation and reconstruction will be available. The next steps will beimproving the event identification algorithms and reconstruction tools and examining physicsperformance for different detector geometries.
Dual-Phase TPC: Simulation and reconstruction for the dual-phase design has been in-corporated into LArSoft and validated. The next steps will be refining the reconstructionalgorithms and doing physics performance studies with the dual-phase geometry.
75
Detector Optimization Studies: Detector optimization studies have started, by firstlooking at the effect of wire angle and spacing on tracking efficiency and e/γ separation. Thenext step will be to see the effect of these choices on the physics sensitivities. The PhotonDetection System Task Force, which is a sub-group of the FD Optimization Task Force, hasjust been formed. They will study the photon detection system requirements with regard tosupernova and nucleon decay physics. Other detector designs, such as perpendicular APAsor TPCs with additional wire planes, are also actively being investigated.
76
References
[1] P. Huber, M. Lindner and W. Winter, Comput. Phys. Commun. 167, 195 (2005)doi:10.1016/j.cpc.2005.01.003 [hep-ph/0407333].
[2] http://www.mpi-hd.mpg.de/lin/globes/
[3] http://larsoft.org/
[4] https://cdcvs.fnal.gov/redmine/projects/larbatch/wiki
[5] http://dune-data.fnal.gov/mc/mcc6.0/index.html
[6] http://www-pnp.physics.ox.ac.uk/~barr/fluxfiles/0401/index.html
[7] http://www.icrr.u-tokyo.ac.jp/~mhonda/nflx2014/index.html
[8] M. Bhattacharya, A. Garcia, N. I. Kaloskamis, E. G. Adelberger, H. E. Swanson, et al.,Phys. Rev. C 58, 3677 (1998) doi:10.1103/PhysRevC.58.3677.
[9] J. A. Cameron and B. Singh, Nuclear Data Sheets, V. 102, I. 2, 293 (2004)doi:10.1016/j.nds.2004.06.001.
[10] http://www.phy.duke.edu/~schol/snowglobes
[11] J. Gava, J. Kneller, C. Volpe, G. C. McLaughlin, Phys. Rev. Lett., 103, 071101 (2009)doi:10.1103/PhysRevLett.103.071101
[12] P. Antonioli, C. Ghetti, E.V. Korolkova, V.A. Kudryavtsev, G. Sartorelli, A Three-dimensional Code for Muon Propagation Through the Rock: MUSIC, AstroparticlePhysics, 7:357, 1997.
[13] V. A. Kudryavtsev, Muon Simulation Codes MUSIC and MUSUN for UndergroundPhysics, Computer Physics Communications, 180:339, 2009.
[14] ArgoNeuT Collaboration, “A study of electron recombination using highly ionizingparticles in the ArgoNeuT Liquid Argon TPC,” JINST 8, P08005 (2013).
[15] ICARUS Collaboration, “Study of electron recombination in liquid argon with theICARUS TPC”, Nucl. Instrum. Meth. Phys. Res. A 523 (2004), 275.
[16] Yichen Li, ”Measurement of Longitudinal Electron Diffusion in Liquid Argon”, Nucl.Instrum. Meth. A816, 160 (2016).
[17] Summary of Liquid Argon Properties http://lar.bnl.gov/properties/.
[18] R. Veenhof, ”GARFIELD, recent developments”, Nucl. Instrm. Meth. Phys. Res. A419 (1998), 726.
[19] Laurence W. Nagel and D. O. Pederson, ”SPICE (Simulation Program iwth IntegratedCircuit Emphasis)”, UCB/ERL M382 (1973).
77
[20] ”A Method to Extract the Charge Distribution Arriving at the TPC Wire Planes inMicroBooNE”, MicroBooNE NOTE 1017-PUB.
[21] D. Whittington, S. Mufson, J. Inst., 11(05), P05016 (2014) doi:10.1088/1748-0221/11/05/P05016 [arXiv:1408.1763 [physics.ins-det]]
[22] ”Noise Characterization and Filtering in the MicroBooNE TPC”, MicroBooNE NOTE1016-PUB.
[23] Bruce Baller, ”Liquid Argon TPC Formation, Signal Processing, and Reconstruction”,under preparation.
[24] https://cdcvs.fnal.gov/redmine/documents/727
[25] http://larsoft.org/single-record/?pdb=110
[26] J. S. Marshall and M. A. Thomson, Eur. Phys. J. C 75, no. 9, 439 (2015)doi:10.1140/epjc/s10052-015-3659-3 [arXiv:1506.05348 [physics.data-an]].
[27] M. Antonello et al., Adv. High Energy Phys. 2013, 260820 (2013)doi:10.1155/2013/260820 [arXiv:1210.5089 [physics.ins-det]].
[28] http://larsoft.org/single-record/?pdb=113
[29] http://www.phy.bnl.gov/wire-cell/
[30] R. Acciarri et al. [DUNE Collaboration], arXiv:1512.06148 [physics.ins-det].
[31] http://www.phy.duke.edu/~schol/snowglobes
[32] S. Amoruso et al. [ICARUS Collaboration], Eur. Phys. J. C 33, 233 (2004)doi:10.1140/epjc/s2004-01597-7 [hep-ex/0311040].
[33] G. Sinev, Talk at LEPP-3, Moscow
[34] G. Sinev, DUNE DocDB-665
[35] Alison J. Roeth, LBNE DocDB-8825
[36] DUNE DocDB-1589
[37] G. Sinev, DUNE DocDB-1181
[38] A. Bueno, Z. Dai, Y. Ge, M. Laffranchi, A. J. Melgarejo, A. Meregaglia, S. Navasand A. Rubbia, JHEP 0704, 041 (2007) doi:10.1088/1126-6708/2007/04/041 [hep-ph/0701101].
[39] C. Andreopoulos et al., Nucl. Instrum. Meth. A 614, 87 (2010)doi:10.1016/j.nima.2009.12.009 [arXiv:0905.2517 [hep-ph]].
[40] C. Andreopoulos, C. Barry, S. Dytman, H. Gallagher, T. Golan, R. Hatcher, G. Perdueand J. Yarba, arXiv:1510.05494 [hep-ph].
78
[41] K. A. Olive et al. [Particle Data Group Collaboration], Chin. Phys. C 38, 090001(2014). doi:10.1088/1674-1137/38/9/090001
[42] M. Sorel, Expanding GENIE’s Nucleon Decay Tools, https://indico.fnal.gov/
getFile.py/access?contribId=4&resId=0&materialId=slides&confId=12549
[43] T. Yang, Simulation of Nucleon Decays, https://indico.fnal.gov/getFile.py/
access?contribId=5&resId=0&materialId=slides&confId=12549
[44] G. Santucci, Far detector simulation and reconstruction chain for non-beamevents, https://indico.fnal.gov/getFile.py/access?contribId=59&sessionId=19&resId=0&materialId=slides&confId=10612
[45] A. Ferrari, P. R. Sala, A. Fasso and J. Ranft, CERN-2005-010, SLAC-R-773, INFN-TC-05-11.
[46] K. Wood, Vertex Assignment Correction in NDK Generation, https:
//indico.fnal.gov/getFile.py/access?contribId=5&resId=0&materialId=
slides&confId=12377
[47] J. Hewes, Neutron-antineutron oscillation in DUNE”, https://indico.fnal.gov/
getFile.py/access?contribId=58&sessionId=31&resId=0&materialId=slides&
confId=10276
[48] A. Higuera, Far detector tracking efficiencies, https://indico.fnal.gov/getFile.py/access?contribId=63&sessionId=19&resId=0&materialId=slides&confId=
10612
[49] H. Mendez and N. Martinez, First look at fully reconstructed p → l+ρ0 andp → µ+K0 events, https://indico.fnal.gov/getFile.py/access?contribId=3&
resId=0&materialId=slides&confId=12549
[50] A. Higuera, Proton decay, p → ν̄K+, https://indico.fnal.gov/getFile.py/
access?contribId=2&resId=0&materialId=slides&confId=12651
[51] R. Acciarri et al. [ArgoNeuT Collaboration], JINST 8, P08005 (2013)doi:10.1088/1748-0221/8/08/P08005 [arXiv:1306.1712 [physics.ins-det]].
[52] K. Wood, Flash Finding Efficiencies for PDK events, https://indico.fnal.gov/
getFile.py/access?contribId=0&resId=0&materialId=slides&confId=12651
[53] M. Goodman and L Lin, Comprehensive list of nucleon decay modes (DUNE-doc-679),http://docs.dunescience.org:8080/cgi-bin/ShowDocument?docid=679
[54] M. Sorel, ndksens, A tool for evaluating NDK sensitivities, https://indico.fnal.gov/getFile.py/access?contribId=57&sessionId=31&resId=0&materialId=
slides&confId=10276
[55] G. D. Barr, T. K. Gaisser, P. Lipari, S. Robbins and T. Stanev, Phys. Rev. D 70,023006 (2004) doi:10.1103/PhysRevD.70.023006 [astro-ph/0403630].
79
[56] G. Battistoni, A. Ferrari, P. Lipari, T. Montaruli, P. R. Sala and T. Rancati, Astropart.Phys. 12, 315 (2000) doi:10.1016/S0927-6505(99)00110-3 [hep-ph/9907408].
[57] M. Sajjad Athar, M. Honda, T. Kajita, K. Kasahara and S. Midorikawa, Phys. Lett.B 718, 1375 (2013) doi:10.1016/j.physletb.2012.12.016 [arXiv:1210.5154 [hep-ph]].
[58] J. Losecco, Honda atmospheric flux calculation for Homestake, https://indico.
fnal.gov/materialDisplay.py?contribId=3&materialId=slides&confId=11533
[59] M. Rafi Alam, I. Ruiz Simo, M. Sajjad Athar and M. J. Vicente Vacas, Phys. Rev. D82, 033001 (2010) doi:10.1103/PhysRevD.82.033001 [arXiv:1004.5484 [hep-ph]].
[60] M. Alam et al., arXiv:1512.06882 [hep-ph].
[61] A. Ioannisian, Exploring the Earth with neutrinos, https://indico.fnal.gov/
materialDisplay.py?contribId=0&materialId=slides&confId=10573
[62] L. Fantini and G. VanWinkle, Measuring the azimuthal angle distribution of at-mospheric neutrinos in DUNE, https://indico.fnal.gov/materialDisplay.py?
contribId=3&materialId=slides&confId=12005
[63] E. Richard et al. [Super-Kamiokande Collaboration], Phys. Rev. D 94, no. 5, 052001(2016) doi:10.1103/PhysRevD.94.052001 [arXiv:1510.08127 [hep-ex]].
[64] H. Gallagher, Atmospheric event generators, https://indico.fnal.gov/
materialDisplay.py?contribId=2&materialId=slides&confId=12274
[65] V. A. Kudryavtsev, Comput. Phys. Commun. 180, 339 (2009)doi:10.1016/j.cpc.2008.10.013 [arXiv:0810.4635 [physics.comp-ph]].
[66] V. A. Kudryavtsev, N. J. C. Spooner and J. E. McMillan, Nucl. Instrum. Meth. A505, 688 (2003) doi:10.1016/S0168-9002(03)00983-5 [hep-ex/0303007].
[67] P. Antonioli, C. Ghetti, E. V. Korolkova, V. A. Kudryavtsev and G. Sartorelli, As-tropart. Phys. 7, 357 (1997) doi:10.1016/S0927-6505(97)00035-2 [hep-ph/9705408].
[68] V. A. Kudryavtsev, E. V. Korolkova and N. J. C. Spooner, Phys. Lett. B 471, 251(1999) doi:10.1016/S0370-2693(99)01378-7 [hep-ph/9911493].
[69] D.-M. Mei, C. Zhang, K. Thomas and F. Gray, Astropart. Phys. 34, 33 (2010)doi:10.1016/j.astropartphys.2010.04.003 [arXiv:0912.0211 [nucl-ex]].
[70] C. Zhao (USD), private communication
[71] F. E. Gray, C. Ruybal, J. Totushek, D.-M. Mei, K. Thomas and C. Zhang, Nucl. In-strum. Meth. A 638, 63 (2011) doi:10.1016/j.nima.2011.02.032 [arXiv:1007.1921 [nucl-ex]].
[72] J. Heise, AIP Conf. Proc. 1604, 331 (2014) doi:10.1063/1.4883449 [arXiv:1401.0861[physics.ins-det]].
80
[73] Martin Richardson, Ph.D. Thesis, University of Sheffield, 2016.
[74] M. L. Cherry, M. Deakyne, K. Lande, C. K. Lee, R. I. Steinberg, B. T. Cleveland andE. J. Fenyves, Phys. Rev. D 27, 1444 (1983). doi:10.1103/PhysRevD.27.1444
[75] N. Abgrall et al. [Majorana Collaboration], arXiv:1602.07742 [nucl-ex].
[76] D. Stefan and A. M. Ankowski, Acta Phys. Polon. B 40, 671 (2009) [arXiv:0811.1892[nucl-th]].
[77] J. Klinger, V. A. Kudryavtsev, M. Richardson and N. J. C. Spooner, Phys. Lett. B746, 44 (2015) doi:10.1016/j.physletb.2015.04.054 [arXiv:1504.06520 [physics.ins-det]].
[78] https://www.overleaf.com/5565604ftcwxj#/17932618/
[79] https://cdcvs.fnal.gov/redmine/projects/larreco/repository/changes/
larreco/TrackFinder/NeutrinoTrackingEff_module.cc
[80] M. Bishai et al. ”LArTPC: Perpendicular vs. Parallel Detector Optimization”, DUNEDocDB-1708.
[81] ”Noise Characterization and Filtering in the MicroBooNE TPC”, MicroBooNE NOTE1016-PUB, http://www-microboone.fnal.gov/publications/publicnotes/
MICROBOONE-NOTE-1016-PUB.pdf.
[82] ”A Method to Extract the Charge Distribution Arriving at the TPC Wire Planes inMicroBooNE”, MicroBooNE NOTE 1017-PUB, http://www-microboone.fnal.gov/publications/publicnotes/MICROBOONE-NOTE-1017-PUB.pdf.
[83] J. Dolph, “LBNF-DUNE Requirements,” DUNE DocDB-112-v6.
81